<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Yevheniia Mala on Medium]]></title>
        <description><![CDATA[Stories by Yevheniia Mala on Medium]]></description>
        <link>https://medium.com/@ymala?source=rss-61c1ec8ecc92------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*0q7TEOvR1b-sWqmm9eHlrg.jpeg</url>
            <title>Stories by Yevheniia Mala on Medium</title>
            <link>https://medium.com/@ymala?source=rss-61c1ec8ecc92------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 01:50:17 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ymala/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Confessions of a Tech Writer]]></title>
            <link>https://medium.com/@ymala/confessions-of-a-tech-writer-03b5b5e64ea0?source=rss-61c1ec8ecc92------2</link>
            <guid isPermaLink="false">https://medium.com/p/03b5b5e64ea0</guid>
            <category><![CDATA[writing-concepts]]></category>
            <category><![CDATA[technical-writing]]></category>
            <category><![CDATA[documentation]]></category>
            <category><![CDATA[docs-as-code]]></category>
            <dc:creator><![CDATA[Yevheniia Mala]]></dc:creator>
            <pubDate>Fri, 03 Jul 2026 17:57:00 GMT</pubDate>
            <atom:updated>2026-07-07T08:18:42.576Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*e7ZSxsmSorF8EjhXKmNcww.png" /></figure><p>Documentation quality depends primarily on the writer. We write with the thought in mind that a particular feature is new and therefore must be explained and published. But who is the reader? From my experience, users don’t read documentation or only read it when something goes wrong. Most likely, they write to customer support instead, and the problem falls on the support team’s shoulders. Documentation is one of the most important tools that help support customers and must be maintained well.</p><p>Historically, documentation or explanatory texts were created in MS Word or other text editors. It was familiar, convenient, and simple. Later, documentation moved online to hosting services like Intercom or HubSpot, which are also convenient thanks to their combination of several features, such as live chat with clients or ticket integration. This is the philosophy that ties documentation closely to customer support. It’s what we want — so why bother changing it?</p><h3>Whole other philosophies</h3><p>If we talk about different approaches in creating and maintaining documentation, we should acknowledge what already exists. In fact, there are more ways than you can imagine of maintaining the documentation, especially within competitive industries, where each detail of the product plays a part in user experience.</p><ul><li><strong>Vendor platforms:</strong> Content lives on third-party servers and is built for support workflows. It may be a SaaS CMS (WordPress, Confluence) or Help Desk.</li><li><strong>Wiki-like:</strong> Community-driven or collaborative browser-based editing that may have the most relevant information, but has weak review processes and versioning.</li><li><strong>Docs-as-Code:</strong> An approach that treats creating documentation as writing software source code. A highly versatile approach that allows creating structured documentation, though fairly demanding for non-technical contributors.</li><li><strong>Structured authoring:</strong> Documentation structure is clear and rigid. The approach allows maximum control over terminology and style, is the most effective for content reuse, but is highly demanding for writers to adopt.</li></ul><p>I’ve worked with most of these — vendor platforms and Docs-as-Code the longest — and each comes with its own version of pain.</p><h3>Hosting services fail you</h3><p>Despite such variety, platforms that provide hosting documentation along with the customer support services may fail you in ways that have never been considered.</p><ul><li><strong>The work that is done twice or more. Much more:</strong> When your product updates one teeny-tiny feature, you go and change one article. Simple, right? But what happens when this is a major update? Your team must check and correct dozens of articles and write the same change over and over again. And there’s no guarantee that they won’t miss anything. Not to mention the time it takes. Will they make it before the release?</li><li><strong>Copy-paste everything:</strong> It’s not a big deal at first: write documentation in MS Word or Google docs, insert the screenshots, save the file or the link, send it for review, then copy everything and paste it into the platform, select the section, and click “Publish”. Then, you find your article with broken images, links and incoherent formatting. You open the document and start fixing the issues that didn’t exist in your original file.</li><li><strong>Many versions? Prepare for the maintenance nightmare:</strong> At some point users stop upgrading the product they use — whether because their license stopped covering the upgrade or they simply decided that the version they use now is the one and only. Then, how should you highlight the difference between the versions? Banner every change! Oh, look, you have 20 different banners in one article already!</li><li><strong>Individuality… What individuality:</strong> Tonight you dreamt of your perfect website: the elegant structure, simple but powerful search engine, a clear path to the most important documents. Will you get it with the platform run by someone else? The answer is “maybe”. The platform may be everything you want, but ultimately you don’t have access to the search engine architecture or the way the content is stored internally. Though you may customize the exterior, the interior is out of reach.</li></ul><p>I can go on and on about the cons. That doesn’t mean that there aren’t any pros or that there are no businesses that benefit from different approaches. I’d be lying if I said there weren’t any.</p><h3>It’s all about the structure</h3><p>Having drowned in the sea of identical changes, I entered into the Docs-as-Code world, and the intentional structure became my main focus. There are different levels of structure to pay attention to:</p><ul><li><strong>Naming</strong> — Can I understand the content from its name?</li><li><strong>Ownership</strong> — Where do I put the article?</li><li><strong>Information</strong> — What information do I have and what should I publish?</li><li><strong>Relationship</strong> — How do documents connect to each other?</li><li><strong>Reuse</strong> — Is there information that appears/will appear in other docs?</li><li><strong>Variants</strong> — How do we represent differences and versions?</li><li><strong>Governance</strong> — Who reviews and approves changes?</li></ul><p>I answered some of these questions once and for all. Others, I had to decide on every time. The most interesting questions were the latter ones.</p><p>The greatest impact on my view of creating documentation was the<strong> Reuse level</strong>. The moment I realized that I don’t have to write the same sentences for the 20th time to describe how to upgrade the product from version x.x to version x+1.x made me tear up a little. That was so simple — just having a file that uses dynamic variables and then setting the value of these variables before including the file into the final document.</p><p>Another example is <strong>Liquid conditionals</strong>. I could set specific data to be displayed on particular pages and keep the rest of the document without changes.<strong> If-then logic</strong> can be applied to the documents too!</p><p>The ways the information can be reused are actually more than that. The images, the banners, next steps, whatever you need. You can create the document by layering the separate components just like a cake. And whenever you have to make a change — you make one and it applies to every document that uses this component.</p><h3>Another kind of engineering</h3><p>Over time, I’ve started to think about building documentation rather than writing it. At first glance, there’s no big difference — you still write words to express what you need to say. However, the true building is in planning, thinking it over, and creating the structure that speaks to the audience. Before writing, one should realize who will read this document. The initial goal for developers and users differs substantially. A technical writer is not just a writer but an engineer who designs a system that supports and reflects the product. Once I realized that, I was able to see outside the box of text editors.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=03b5b5e64ea0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The IoT Protocol Breakdown Nobody Finishes]]></title>
            <link>https://medium.com/@ymala/the-iot-protocol-breakdown-nobody-finishes-438e0ea165f4?source=rss-61c1ec8ecc92------2</link>
            <guid isPermaLink="false">https://medium.com/p/438e0ea165f4</guid>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[iot]]></category>
            <category><![CDATA[api-gateway]]></category>
            <category><![CDATA[edge-computing]]></category>
            <category><![CDATA[network-protocols]]></category>
            <dc:creator><![CDATA[Yevheniia Mala]]></dc:creator>
            <pubDate>Thu, 19 Mar 2026 14:55:10 GMT</pubDate>
            <atom:updated>2026-03-19T14:55:10.490Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EWn3c-NPjNGSMiDpEX7qfg.png" /></figure><p>There are now roughly <strong>21 billion connected IoT devices</strong> in the world, and that number is growing at around 14% a year — putting us on track for 39 billion by 2030, according to<a href="https://iot-analytics.com/number-connected-iot-devices/"> IoT Analytics</a>. That’s a lot of sensors, actuators, meters, cameras, and trackers, all generating data that needs to go somewhere.</p><p>Most of those devices are still connecting to the cloud by default. This works fine when you have a handful of devices on a reliable connection. It starts to fall apart when you’re managing hundreds of sensors on a LoRaWAN network, legacy Modbus equipment in a factory basement, or a mix of five different protocols across three buildings.</p><p>Protocols were never the problem. So what is?</p><h3>First, a quick map of the landscape</h3><p>IoT protocols as of 2026 aren’t a single spectrum. They divide roughly into three families, each making fundamentally different trade-offs.</p><p><strong>Short-range, high-bandwidth</strong>:<strong> Wi-Fi, BLE, UWB.</strong></p><p>These are your workhorses for environments where power isn’t a critical constraint and you need actual data throughput. Wi-Fi accounts for about 32% of all IoT connections globally; Bluetooth another 24%. Together they carry nearly two-thirds of the market. That share is earned: if you’re running a security camera, a smart display, or any device with a permanent power source and a local network, there’s rarely a good reason to use anything else.</p><p>UWB deserves a separate mention. It’s not a high-bandwidth protocol in the usual sense — it’s a precision <em>ranging</em> technology. By measuring the time-of-flight of radio pulses across 500 MHz-wide channels, it can locate a device to within 10–30 cm. The<a href="https://www.firaconsortium.org/news/press-releases/2025/01/fira-consortium-unveils-fira-core-3-0-specifications-and-certification"> FiRa Core 3.0 specification</a>, released in January 2025, added dedicated data transfer and improved scheduling for simultaneous ranging and data exchange — opening up use cases like tap-free transit payment and secure car access that need both location and data in milliseconds. This is meaningfully different from what you get from BLE RSSI positioning, which measures signal strength and can be spoofed with a relay attack. UWB can’t be.</p><p><strong>Low-power wide-area (LPWAN): LoRaWAN, NB-IoT, LTE-M, Sigfox.</strong></p><p>These trade throughput for range and battery life. A LoRaWAN sensor in a rural field can run on a single battery for a decade and still hit a gateway 15 km away. NB-IoT uses licensed spectrum and existing cellular infrastructure, which gives it predictable indoor penetration — its maximum coupling loss of 164 dB makes it the right call for meters in basements and underground manholes. LTE-M sits in between: it supports handover between cell towers (unlike NB-IoT), which makes it the sensible choice for anything that moves — fleet telematics, wearable medical devices, mobile assets.</p><p>The choice between LoRaWAN and NB-IoT comes down less to specs and more to ownership model. LoRaWAN lets you run a private network — your gateways, your data. NB-IoT connects through carrier infrastructure, which means nationwide coverage out of the box but a monthly SIM cost per device that compounds fast at scale.</p><p><strong>Mesh protocols for smart buildings: Thread, Zigbee, Z-Wave.</strong></p><p>Thread is the most interesting development here. It’s an IPv6 mesh protocol that runs on the same 802.15.4 radio as Zigbee but gives every device its own IP address and supports self-healing mesh topology. The Matter standard — backed by Apple, Google, Amazon, and others — runs on top of Thread (and Wi-Fi and Ethernet), which is why Thread has gone from niche to default for new battery-powered smart building devices essentially overnight. Zigbee isn’t going anywhere. There are hundreds of millions of nodes deployed, but new installations are increasingly Thread-first.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YcD7nd2P8SbSedteSq4k5Q.png" /><figcaption>Protocol comparison table</figcaption></figure><h3>All these protocols land at the same place</h3><p>Here’s where it gets interesting. You can design a perfectly rational multi-protocol deployment — LoRaWAN sensors in the field, NB-IoT meters underground, Modbus actuators in the control room, BLE beacons for indoor tracking — and then hit a wall when all that data has to go somewhere useful.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Wdq7LMZVdZKGBOAoJ73glA.png" /><figcaption><em>Two approaches to IoT data routing</em></figcaption></figure><p>In the default architecture, every device sends raw data to the cloud. The cloud processes it, applies rules, maybe sends a command back. This is fine until:</p><ul><li><strong>Latency matters.</strong> If a vibration sensor detects a bearing failure pattern, you don’t want to wait for a round trip to a data center before triggering an alert. A few hundred milliseconds of network jitter turns a predictive maintenance system into a reactive one.</li><li><strong>Connectivity is unreliable.</strong> A LoRaWAN gateway on a factory roof may have solid uptime. The internet connection at that facility? Not necessarily. If your pipeline breaks when WAN goes down, you don’t have an edge deployment — you have a cloud deployment with an optimistic assumption about connectivity.</li><li><strong>Volume becomes a cost problem.</strong> IDC forecast back in 2019 that IoT devices would generate approximately<a href="https://www.businesswire.com/news/home/20190618005012/en/The-Growth-in-Connected-IoT-Devices-is-Expected-to-Generate-79.4ZB-of-Data-in-2025-According-to-a-New-IDC-Forecast"> 79.4 zettabytes of data annually by 2025</a>. Whether or not that forecast was accurate down to the digit, the directional reality is obvious to anyone who’s looked at a cloud bill for a high-volume deployment. Video surveillance data, high-frequency sensor polling, protocol bridges that forward every packet upstream. It all adds up fast.</li><li><strong>Protocol translation needs to happen somewhere.</strong> If you have Modbus RS-485 equipment that needs to talk to an MQTT broker, something has to convert it. That something is a gateway. But a gateway that only translates and then forwards everything to the cloud is just creating a bottleneck.</li></ul><h3>The gateway is where the protocol conversation should actually end</h3><p>There’s a tendency in IoT architecture diagrams to draw gateways just as dumb boxes that aggregate traffic from the edge and hand it off upstream. That’s historically where most commercial gateways stopped. The more useful framing is to treat the gateway as the first intelligent layer in your stack.</p><p>What does intelligence at the gateway actually mean in practice?</p><p><strong>Filtering and pre-processing.</strong> A temperature sensor that reports every second doesn’t need to send every reading to the cloud. If the value is within normal range and hasn’t changed meaningfully, the gateway can suppress it. Only the delta, the anomaly, or the summary goes upstream. This alone can cut data volume substantially without losing any analytically useful signal.</p><p><strong>Local rule execution.</strong> The rule that says “if pressure exceeds X, close valve Y” doesn’t need to travel to the cloud and back. An edge runtime on the gateway keeps the control loop tight regardless of WAN status.</p><p><strong>Protocol normalization.</strong> A modern gateway handles Modbus, OPC-UA, HTTP, and often custom serial protocols under one roof. The data model gets normalized at the edge, and by the time it reaches the cloud, it’s already structured and labeled</p><p><strong>Offline operation.</strong> The ability to buffer and keep processing during WAN outages is what separates a gateway from a pipe. The distinction matters when you’re deploying in locations where connectivity is opportunistic rather than guaranteed.</p><h3>What this looks like in practice</h3><p>Take a manufacturing site with a mix of legacy Modbus sensors, a handful of LoRaWAN nodes in the yard, and a requirement to trigger local alarms when certain thresholds are crossed. In a cloud-first setup, all of that data travels upstream raw — the rule evaluation happens in the platform, and if the WAN connection drops, the alarms stop firing.</p><p>An edge-capable gateway changes that equation. Protocol translation happens on-site. Rules execute locally. Data gets buffered during outages and synchronized upstream when connectivity restores. The cloud still gets a full picture — it’s just not in the critical path for anything time-sensitive.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tZbcM94Hv6bDDoubPoWcFA.png" /><figcaption>Edge-first architecture: local processing, selective upstream sync</figcaption></figure><p>Platforms like <a href="https://thingsboard.io/products/thingsboard-edge/?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=protocol-breakdown">ThingsBoard Edge</a> take this further by running a full local instance — rule engine, dashboards, data persistence — on the gateway hardware itself. Local operators can monitor and interact with the system without depending on a cloud connection. It’s one approach among several, but it shows concretely what the difference between a forwarder and an edge platform looks like from the inside.</p><h3>So what should drive your protocol decisions?</h3><p>A few practical questions that tend to cut through the spec-sheet comparisons:</p><p><strong>How long does the device need to run on battery, and how often does it need to transmit?</strong> If the answer is “years” and “occasionally,” you’re in LoRaWAN or NB-IoT territory. If the answer is “it’s plugged in” or “continuously,” Wi-Fi or LTE-M makes more sense.</p><p><strong>Does the data need to trigger a real-time response?</strong> If a sensor reading should cause an action within seconds, that logic needs to run close to the hardware — either on the device itself or on a gateway with local processing. Cloud-dependent control loops are latency-dependent by definition.</p><p><strong>Who owns the network?</strong> For industrial deployments, private LoRaWAN networks often make financial and operational sense — you control the gateways, you control the data, there are no per-device SIM costs. For deployments that need to span city-scale geography without deploying infrastructure, NB-IoT or LTE-M on carrier networks is the pragmatic answer.</p><p><strong>What’s already deployed?</strong> This is the one that gets skipped in greenfield discussions. Most industrial environments have years of Modbus, RS-485, or proprietary serial equipment that isn’t going anywhere. Protocol selection often isn’t a clean decision — it’s a translation problem. Whatever sits between the legacy layer and the modern stack needs to handle both.</p><h3>The uncomfortable truth about “just send it to the cloud”</h3><p>The default architecture — every device sends everything to a central platform — made sense when IoT deployments were small, when bandwidth was cheap, and when the processing requirements were simple. None of those assumptions hold at scale.</p><p>The protocols at the bottom of your stack are mature. Wi-Fi, LoRaWAN, NB-IoT, Thread — they all do what they claim to do. The gap is usually in what happens between the sensor and the cloud: how data gets aggregated, filtered, translated, and acted on before it travels the last mile upstream.</p><p>Getting that layer right is, increasingly, the real work.</p><p><strong>Where to go next</strong></p><ul><li>Need software-based protocol translation? → <a href="https://thingsboard.io/docs/iot-gateway/what-is-iot-gateway/?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=protocol-breakdown">ThingsBoard IoT Gateway</a></li><li>Evaluating edge platforms? →<a href="https://thingsboard.io/docs/edge/getting-started-guides/what-is-edge/?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=protocol-breakdown"> ThingsBoard Edge</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=438e0ea165f4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[When Cloud Latency Fails: How Edge Computing Enables Real-Time Grid Control]]></title>
            <link>https://medium.com/@ymala/when-cloud-latency-fails-how-edge-computing-enables-real-time-grid-control-bbb54f006b06?source=rss-61c1ec8ecc92------2</link>
            <guid isPermaLink="false">https://medium.com/p/bbb54f006b06</guid>
            <category><![CDATA[edge-computing]]></category>
            <category><![CDATA[water-utility]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[power-grid]]></category>
            <dc:creator><![CDATA[Yevheniia Mala]]></dc:creator>
            <pubDate>Wed, 18 Feb 2026 14:01:02 GMT</pubDate>
            <atom:updated>2026-02-19T15:02:58.995Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*at8Sz5zCVbep4jkrQbogiw.png" /></figure><p>The power grid is undergoing its most significant architectural shift since electrification. For decades, the model was simple: generate power at large facilities, transmit it in one direction to consumers, collect sparse data for billing. Intelligence lived in central control rooms processing aggregated information.</p><p>Cloud computing was a perfect fit for that model, but now it’s straining under three converging pressures: the proliferation of distributed energy resources that make the grid bidirectional and volatile, climate-driven demands for water conservation requiring precise leak detection, and an explosion of high-resolution sensor data that centralized systems can’t process economically or fast enough.</p><p>Round-trip latency to cloud services varies significantly based on network conditions and geographic distance to data centers. Well-optimized public cloud connections typically achieve 50–100 milliseconds, though latency can exceed 200 milliseconds during network congestion or when data traverses multiple regions. For historical analysis, billing reconciliation, and monthly reporting, this performance is adequate. For time-sensitive grid operations, such latency becomes problematic. Cloud round-trip latency means commands arrive after conditions have shifted.</p><p>The bandwidth economics are equally problematic. Modern sensors generate torrents of data. A utility deploying 100,000 advanced meters capturing high-resolution waveforms can produce tens to hundreds of terabytes daily, depending on sampling rates. Transmitting everything to the cloud becomes prohibitively expensive. Utilities are forced to downsample or aggregate, losing the granularity needed for sophisticated control.</p><h3>Edge Architecture Principles</h3><p>The logical response to these obstacles is to move intelligence from data centers to the periphery where data originates. <a href="https://thingsboard.io/docs/edge/getting-started-guides/what-is-edge/?fpr=ai432j&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=When+Cloud+Latency+Fails">Edge computing</a> addresses these failures through three core principles: local processing, autonomous operation, and bidirectional synchronization.</p><p><strong>Local processing </strong>means running analytics, AI models, and control logic where data originates. This eliminates round-trip latency and reduces bandwidth consumption by orders of magnitude. Only insights and events transmit to central systems, not raw telemetry streams.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pdkZKCVL3YRPoSevx-Qbhg.png" /><figcaption>Local processing</figcaption></figure><p><strong>Autonomous operation </strong>ensures edge devices continue functioning during connectivity loss. If the central system goes offline, distributed intelligence maintains grid stability, controls DERs, and monitors infrastructure until connectivity restores.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vpLwzZ1yLLLzvtXUqKmhXw.png" /><figcaption>Autonomous operation</figcaption></figure><p><strong>Bidirectional synchronization</strong> means edge devices aren’t isolated islands. They sync state with central systems when connectivity is available, enabling fleet-wide visibility and coordinated control while maintaining local autonomy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0_Zf6o6fDX1GsrkYDBun7A.png" /><figcaption>Bidirectional synch</figcaption></figure><p>These principles require a platform that operates across the entire stack. Platforms like ThingsBoard Edge provide this foundation while maintaining local autonomy when connectivity is lost.</p><h3>Implementing Smart Meter Edge Intelligence</h3><p>Perhaps the most visible edge computing evolution is happening at the electric meter itself. Utilidata partnered with NVIDIA to embed AI directly into meters using the Jetson platform. These devices run complex optimization algorithms locally, analyzing high-resolution voltage and current waveforms in real-time.</p><p>In a May 2025 <a href="https://www.epri.com/research/products/000000003002031878">demonstration project with Southern California Edison and EPRI</a>, Utilidata’s Karman platform tested real-time distributed energy resource coordination. The system performed Real-Time Optimal Power Flow calculations at the meter. Solar inverters, EV chargers, and battery systems received control signals based on instantaneous local grid conditions rather than pre-scheduled dispatch from central systems.</p><p>The demonstration achieved a 27% reduction in peak demand and 12.5% reduction in electricity costs for the simulated home. More significantly, the meters detected power quality anomalies, before they caused outages, enabling proactive maintenance.</p><h3>Building Similar Capabilities</h3><p>Utilidata’s demonstration validates the edge computing model for smart meter intelligence. The architectural requirements are clear: utilities need platforms that can collect high-frequency data from diverse meter types, process it through configurable local logic, and manage thousands of distributed edge devices from a central operations center. The platforms must also be able to maintain autonomous operation during connectivity loss.</p><p>For utilities evaluating implementation paths, open-source edge platforms like <a href="https://thingsboard.io/products/thingsboard-edge/?fpr=ai432j&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=When+Cloud+Latency+Fails">ThingsBoard Edge</a> provide this infrastructure. The platform addresses the protocol integration challenge through IoT Gateways that connect Modbus, and other proprietary meter protocols into a unified data stream. The rule engine processes telemetry locally: detecting voltage anomalies, calculating power quality metrics, triggering DER control commands.</p><p>The platform operates autonomously during connectivity loss. If a substation loses network access during a storm, edge gateways continue executing local control rules. When connectivity restores, the edge instance syncs state with the central ThingsBoard server, providing operators fleet-wide visibility.</p><p>The economic model differs from cloud platforms: processing happens locally with no per-message fees. Compute costs are fixed to deployed hardware, not data volume processed, which is significant for utilities generating millions of telemetry points daily.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PTqZBnNJjf_bCJV9C5dLkg.png" /><figcaption>Implementing Electric Utilities Networks</figcaption></figure><h3>Water Utilities: Edge AI Detects the Invisible</h3><p>Another noticeable edge computing and AI revolution happens at water utilities. Water loss represents a massive but largely invisible operational drain. Traditional detection methods rely on manual acoustic surveys or satellite imagery, both expensive and incomplete.</p><p>EPCOR, Arizona’s largest private water utility, <a href="https://news.microsoft.com/source/features/sustainability/ai-tool-uses-sound-to-pinpoint-leaky-pipes-saving-precious-drinking-water/">deployed FIDO Tech’s AI-powered acoustic sensors</a> across San Tan Valley’s 160-square-mile service area. The technology moves beyond simple leak detection to leak classification and prioritization through edge-based AI.</p><p>Deep learning algorithms analyze sound frequency profiles and textures locally at the sensor. The system distinguishes a water leak’s specific hiss from traffic rumble, transformer hum, or pump vibration. Critically, it estimates leak volume, allowing utilities to prioritize repairs. A small weep can be monitored; a large fracture gets immediate attention.</p><p>Over one year, EPCOR identified 250+ leaks. One medium-sized hydrant leak was losing 3–7 gallons per minute. According to Microsoft, this volume was equivalent to the average daily water consumption of 43 households. The leak had no visible indicators — water seeped through the hydrant base, where it would have continued for years until the next maintenance cycle.</p><p>The breakthrough extends to plastic pipes, historically the hardest to monitor. PVC doesn’t resonate like metal, dampening acoustic signals rapidly. Edge-deployed AI models, trained on extensive plastic pipe leak datasets, identify the subtle low-frequency signatures that threshold-based sensors and human listeners miss.</p><p>Impact: EPCOR reduced non-revenue water from 27% to approximately 10%, <a href="http://azcapitoltimes.com/news/2025/10/10/building-water-resilience-through-innovation-and-partnerships/">recovering over 115 million gallons annually</a>.</p><h3>Implementing Distributed Acoustic Sensor Networks</h3><p>EPCOR’s deployment demonstrates that edge-based AI can solve water infrastructure challenges traditional methods couldn’t address. Detecting leaks across 160 square miles of desert infrastructure means deploying hundreds of battery-powered sensors in locations without reliable connectivity, no access to continuous power, and no tolerance for bandwidth-heavy data streams.</p><p>For utilities building similar capabilities, the architecture starts at the sensor level. Edge gateways connect to acoustic sensors via protocols designed for long range, low power, intermittent connectivity — LoRaWAN or NB-IoT. Sensors wake periodically, capture acoustic samples, and process them locally. When a leak signature matches trained patterns, the sensor transmits. When it doesn’t, it goes back to sleep. Continuous audio streaming over cellular networks is economically unfeasible.</p><p>Platforms like ThingsBoard Edge provide the processing layer between sensors and cloud. The edge instance runs AI inference locally, either calling external models via the <a href="https://thingsboard.io/docs/user-guide/rule-engine-2-0/nodes/external/ai-request/?fpr=ai432j&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=When+Cloud+Latency+Fails">AI Request node</a> or executing lightweight models directly on the gateway. When a leak signature is detected, the system creates an alarm with GPS coordinates, estimated flow rate, and urgency classification. Maintenance crews receive work orders prioritized by leak severity, not discovery order.</p><p>Keeping those AI models current across a distributed fleet introduces its own operational challenge. Utilities can’t manually update inference logic on hundreds of field-deployed sensors. ThingsBoard addresses this through centralized model repositories and over-the-air deployment pipelines. Engineers train models on historical data, validate performance, then push updates to device fleets. Edge gateways download new models during scheduled maintenance windows, switching to updated inference logic without interrupting operations.</p><p>Asset management becomes equally critical at this scale. Five hundred sensors across a service area means five hundred batteries to monitor, five hundred communication windows to track, five hundred health statuses to interpret. ThingsBoard’s centralized dashboard provides this fleet-wide visibility while edge nodes maintain autonomous detection regardless of cloud connectivity.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bEwQXHLJuzQPb-jKFvjqcQ.png" /><figcaption>Implementing Distributed Acoustic Sensor Networks</figcaption></figure><h3>Making Edge Computing Work in Harsh Environments</h3><p>The case studies above demonstrate what’s possible. Utility infrastructure operates in conditions cloud-centric architectures weren’t designed to handle: smart meters must maintain grid control during storms that sever connectivity, acoustic sensors buried in desert soil operate for years on battery power in extreme temperatures. Implementing these systems requires understanding the hardware, integration patterns, and security models that make edge computing viable in utility environments.</p><p><strong>Hardware resilience becomes critical.</strong> Edge devices endure temperature extremes, voltage fluctuations, and years between maintenance. Industrial-grade components, fanless designs, and wide-temperature-range processors eliminate frequent replacement costs.</p><p><strong>Power constraints shape architecture.</strong> Edge rule engine processes information locally, activating transmission only when anomalies are detected. This extends the battery life from weeks to years while maintaining detection accuracy.</p><p><strong>Intermittent connectivity requires autonomy.</strong> When storms isolate substations, edge devices continue executing local control rules. Smart meters balance DERs using real-time measurements, not stale cloud instructions. Connectivity restoration triggers automatic state synchronization.</p><p><strong>Security and compliance intensify at the edge.</strong> Processing sensitive infrastructure data locally addresses both attack surface and regulatory concerns. Acoustic streams remain on-premise, and only anonymized insights transmit to central systems, satisfying data sovereignty requirements while maintaining operational visibility.</p><h3>The Path Forward</h3><p>The utilities mastering edge computing share common characteristics: they started with high-value use cases where real-time processing delivers outsized returns, they piloted before scaling to validate ROI and refine algorithms, and they chose open platforms to avoid vendor lock-in and maintain flexibility as requirements evolve.</p><p>The transition from centralized cloud intelligence to distributed edge processing isn’t optional. It’s the only architecture capable of managing the complexity ahead — grids with millions of distributed energy resources, water networks with real-time leak detection, and industrial assets that predict their own maintenance needs.</p><p>The question isn’t whether utilities will adopt edge computing. It’s whether they’ll do so proactively with architectures that scale, or reactively after centralized systems buckle under operational demands.</p><p><strong>Learn how ThingsBoard supports distributed energy infrastructure:</strong> <a href="https://thingsboard.io/case-studies/energenix/?fpr=ai432j&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=When+Cloud+Latency+Fails">Energenix Solar SCADA Case Study</a></p><p><strong>Explore water infrastructure monitoring:</strong> <a href="https://thingsboard.io/use-cases/water-metering/?fpr=ai432j&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=When+Cloud+Latency+Fails">Water Metering Solutions</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bbb54f006b06" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[When Predictions Meet Reality: What IoT Actually Looked Like in 2025]]></title>
            <link>https://medium.com/@ymala/when-predictions-meet-reality-what-iot-actually-looked-like-in-2025-1aaeca9fbb65?source=rss-61c1ec8ecc92------2</link>
            <guid isPermaLink="false">https://medium.com/p/1aaeca9fbb65</guid>
            <category><![CDATA[smart-cities]]></category>
            <category><![CDATA[iot]]></category>
            <category><![CDATA[5g]]></category>
            <category><![CDATA[tech-trends]]></category>
            <category><![CDATA[edge-computing]]></category>
            <dc:creator><![CDATA[Yevheniia Mala]]></dc:creator>
            <pubDate>Tue, 06 Jan 2026 14:02:28 GMT</pubDate>
            <atom:updated>2026-01-06T14:02:28.780Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/455/1*K8CNX5Ry7YAiE5hYd_lnUg.png" /></figure><p>Remember all those breathless predictions about 2025? The year when smart cities would blanket the globe, 5G would transform everything it touched, and IoT security would finally be “solved”?</p><p>Well, 2025 is ending, and while the IoT industry has certainly evolved, the gap between what the experts promised and what actually happened is… let’s just say it’s educational.</p><p>I’ve spent the past few weeks digging through analyst reports, industry data, and real-world deployments to see how those confident forecasts held up. Spoiler alert: reality had other plans.</p><h3>The 5G Revolution That Wasn’t (Quite)</h3><p>Back in 2019, everyone was convinced 5G would completely reshape IoT by 2025. The promise was tantalizing: ultra-low latency, massive device density, and use cases we could barely imagine.</p><p>Here’s what actually happened: According to ITU data, 5G networks now cover about <a href="https://www.gsma.com/somic/wp-content/uploads/2025/09/The-State-of-Mobile-Internet-Connectivity-2025-Network-Coverage-and-Infrastructure.pdf">54%</a> of the world’s population. That sounds impressive until you realize that only <a href="https://techxplore.com/news/2025-11-global-internet-users-disparities-deepen.html">4%</a> of people in low-income countries have 5G access, compared to <a href="https://techxplore.com/news/2025-11-global-internet-users-disparities-deepen.html">84%</a> in high-income nations.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/730/1*_Zf-VpVVYS7IOXoPM3F_mA.png" /><figcaption><em>Figure 1. 5G coverage by income level in 2025, based on ITU data.</em></figcaption></figure><p>But here’s the kicker — most of that coverage is what’s called Non-Standalone (NSA) 5G, which still relies on 4G infrastructure. The truly revolutionary Standalone (SA) networks? By mid-2025, only <a href="https://gsacom.com/paper/5g-standalone-august-2025/">77 operators across 43 markets </a>were operating SA to some degree on their public networks, according to the Global Mobile Suppliers Association.</p><p>For most IoT deployments I’ve seen this year, companies are still choosing 4G, Wi-Fi, or LoRaWAN. Why? They’re cheaper, more widely available, and frankly, they get the job done. The remote surgery and autonomous vehicle scenarios that needed 5G’s unique capabilities? They’re happening, but they remain niche applications rather than mainstream deployments.</p><p>The lesson here isn’t that 5G failed. It’s that infrastructure transformations take way longer than PowerPoint presentations suggest. We keep forgetting that “good enough” technology has remarkable staying power when it’s already paid for and working.</p><h3>Edge Computing: The Both Revolution</h3><p>Remember when edge computing was supposed to kill the cloud? Analysts were throwing around some wild numbers. <a href="https://www.gartner.com/smarterwithgartner/what-edge-computing-means-for-infrastructure-and-operations-leaders">Gartner</a> predicted that 75% of enterprise data would be processed at the edge by 2025, up from just 10% in 2018.</p><p>That prediction wasn’t entirely wrong, but it completely missed the point. Edge didn’t replace cloud computing, it married it.</p><p>What we’re seeing in 2025 is a hybrid approach that nobody really predicted. Companies are using edge for time-sensitive processing and data filtering, then sending the important stuff to the cloud for heavy-duty analytics and long-term storage. It’s not an either/or situation; it’s both.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-9MqeW_ZyLA4MneVRYeD3A.png" /><figcaption><em>Figure 2. Edge + Cloud: not versus, but pipeline</em></figcaption></figure><p>Take manufacturing plants, for instance. They’re processing sensor data locally to catch equipment failures in real-time, but they’re still using cloud platforms for supply chain optimization and predictive maintenance models that need historical data from multiple facilities.</p><p>The binary thinking of “X will replace Y” turned out to be completely wrong. Technology adoption is additive, not substitutional. We layer new capabilities on top of existing ones, creating these complex but powerful hybrid architectures.</p><h3>The Security Problem That Won’t Go Away</h3><p>This one hurts to write about. Every year since about 2015, someone has declared that “this will be the year IoT security gets fixed.”</p><p>In 2025? Not even close.</p><p>Here’s the sobering reality: Between January and October 2025, <a href="https://www.bitdefender.com/en-us/news/new-bitdefender-and-netgear-report-reveals-rising-threats-across-the-connected-home-landscape">Bitdefender</a> detected 13.6 billion attacks against consumer IoT devices. Let that sink in. Billions of attacks, and that’s just what one security company saw.</p><p>The fundamental problems remain stubbornly unchanged. <a href="https://jumpcloud.com/blog/iot-security-risks-stats-and-trends-to-know-in-2025">One in five</a> IoT devices still ships with default passwords. Legacy devices deployed years ago remain unpatched and unpatchable. And despite new regulations like the UK’s Product Security Act (which went into effect in 2024), the economic incentives still favor speed-to-market over security.</p><p>According to <a href="https://www.nozominetworks.com/ot-iot-cybersecurity-trends-insights-july-2025">Nozomi Networks’ 2025 data</a>, 7.36% of detected IoT attacks involve brute-force attempts, while 5.27% exploit default credentials. These aren’t sophisticated nation-state attacks, they’re the digital equivalent of checking if you left your front door unlocked.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BcN98WPhKgt5gXXzsfTLgA.png" /><figcaption><em>Figure 3. Top 10 Most Common MITRE ATT&amp;CK® Techniques Associated with Raised Alerts, based on the Nozomi Networks 2025 report.</em></figcaption></figure><p>The hard truth? Security isn’t a destination you arrive at; it’s an ongoing battle where the defenders are always playing catch-up. And with billions of vulnerable devices already in the wild, we’re dealing with technical debt that will take decades to pay off.</p><h3>AI at the Edge: Physics vs. Marketing</h3><p>“Every device will have AI by 2025!” How many times did we hear that one?</p><p>The reality is more nuanced. Yes, we’re seeing AI capabilities in high-end devices — smartphones, security cameras, industrial gateways. But the vast majority of battery-powered sensors? They’re as “dumb” as ever, and for good reason.</p><p>Running meaningful AI models requires processing power that would drain a sensor battery in hours. The economics just don’t work for devices where every cent of cost and every milliamp of power consumption matters.</p><p>What we’re actually seeing is AI concentration in gateway devices that can afford the power budget, with simpler devices feeding them data. It’s a hub-and-spoke model, not the fully distributed intelligence that was promised.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/717/1*QrWfIoQsaxYLJ3ehfQ8orw.png" /><figcaption><em>Figure 4. The reality: a hub-and-spoke model</em></figcaption></figure><h3>Smart Cities: Still More Pilot Than Production</h3><p>Perhaps no prediction has aged worse than the smart city revolution. The numbers tell the story: The 2025 IMD Smart City Index tracks <a href="https://www.imd.org/smart-city-observatory/home/">146 cities</a> globally, but comprehensive, city-wide smart deployments? They remain frustratingly rare.</p><p>Yes, we’re seeing progress. Cities are deploying smart streetlights, intelligent traffic systems, and connected waste management. But these are typically isolated projects, not the integrated, city-wide platforms we were promised.</p><p>The challenge isn’t technology, it’s everything else. Budget cycles, procurement processes, departmental silos, and the simple fact that cities need to keep the lights on while they’re supposedly transforming. Cities are getting smarter, but they’re doing it one project at a time, not through grand transformations.</p><h3>The Interoperability Dream Deferred</h3><p>And then there’s <a href="https://en.wikipedia.org/wiki/Matter_(standard)">Matter</a>, the smart home standard that was supposed to finally make everything work together. Launched in late 2022 with backing from Apple, Google, Amazon, and others, it promised to be the universal translator for smart homes.</p><p>By 2025? <a href="https://www.gearbrain.com/smart-home-compatibility-gearbrain-2672559280.html">Users are still confused</a> about what Matter even does and whether it actually improves their experience. Legacy devices dominate households and don’t support Matter, while varying manufacturer support means some Matter-enabled devices perform differently depending on brand or firmware.</p><p>The pattern is familiar: new standards don’t replace old ones; they just add to the pile. Instead of solving fragmentation, we’ve added another layer to navigate.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/717/1*0PVA_KvuMc9U3dYXdSy5GA.png" /><figcaption><em>Figure 5. Isolated pilot projects.</em></figcaption></figure><h3>So What Did 2025 Actually Teach Us?</h3><p>After reviewing all these predictions against reality, some clear patterns emerge:</p><p><strong>Infrastructure takes forever.</strong> Whether it’s 5G networks or smart city deployments, physical infrastructure changes happen in decades, not years. The PowerPoint might say 2025, but reality says 2035.</p><p><strong>Economics beats technology.</strong> Many things are technically possible but not economically viable. The best technology doesn’t win; the most cost-effective one does.</p><p><strong>Hybrid wins over binary.</strong> Every “X will replace Y” prediction failed. The answer is almost always “both, depending on context.”</p><p><strong>Security is a process, not a product.</strong> There’s no finish line for security, just an endless race where the stakes keep getting higher.</p><p><strong>Legacy systems have nine lives.</strong> That ancient system everyone says will be replaced? It’ll probably outlive us all.</p><h3>The Value in Being Wrong</h3><p>Should we just ignore predictions then? Actually, no.</p><p>The value isn’t in the predictions themselves — it’s in understanding why they missed the mark. Every wrong forecast teaches us something about how technology actually gets adopted, how markets really work, and what constraints really matter.</p><p>The experts weren’t wrong about the direction of travel. IoT is growing, 5G is rolling out, edge computing is important, and security matters more than ever. They were just wildly optimistic about the timeline and way too simplistic about the path.</p><p>2025’s reality might not match the predictions, but it’s still pretty remarkable. We have billions of connected devices, hybrid cloud-edge architectures, and slowly-but-surely improving security. It’s messier than the vision, but it’s real and it works.</p><p>As we head toward 2026, we’ll undoubtedly see a new batch of predictions. Some will be right, most will be wrong, and all of them will be entertaining to look back on in five years. The cycle continues, and honestly, that’s part of the fun.</p><p>The future arrives eventually. It just takes longer than the keynote presentations suggest, and it’s always messier than the diagrams. But that’s what makes it interesting.</p><p><strong>Building IoT Solutions in the Real World?</strong></p><p>While we wait for universal standards and bulletproof security, real IoT projects still need to ship. See how ThingsBoard Edge handles the messy reality of hybrid deployments, multiple protocols, and legacy devices.</p><p><a href="https://thingsboard.io/docs/edge/getting-started/?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=reality_check_2025"><strong>Explore ThingsBoard Edge →</strong></a></p><p>Open source. Battle-tested. Built for reality, not PowerPoints.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1aaeca9fbb65" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What You Need to Deploy Edge ML in IoT]]></title>
            <link>https://medium.com/@ymala/what-you-need-to-deploy-edge-ml-in-iot-c803e62695c4?source=rss-61c1ec8ecc92------2</link>
            <guid isPermaLink="false">https://medium.com/p/c803e62695c4</guid>
            <category><![CDATA[iot]]></category>
            <category><![CDATA[edge-computing]]></category>
            <category><![CDATA[devops]]></category>
            <category><![CDATA[ml-model-deployment]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Yevheniia Mala]]></dc:creator>
            <pubDate>Thu, 11 Dec 2025 14:42:14 GMT</pubDate>
            <atom:updated>2025-12-11T15:05:48.466Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pqOxsfM4ZXtWzOhYDgHvoQ.png" /></figure><p>Most articles jump straight from “ML is amazing” to “here’s our product.” But between deciding you need edge ML and actually deploying it, there are four foundational pieces you need to get right. Miss any one of them, and your deployment will struggle. Get them working together, and you’ve got a system that actually delivers value.</p><p>Let me walk you through what matters for each component.</p><h3>Edge Hardware: The Foundation</h3><p>Edge ML runs on local devices at your sites, not in cloud data centers. That much is obvious. What’s less obvious is figuring out what hardware you actually need.</p><p>Hardware requirements depend entirely on your model complexity and deployment environment. The challenge is that there’s no one-size-fits-all answer here. Your deployment environment, model complexity, and budget all factor in. This is why talking to system integrators who know your industry makes more sense than trying to follow generic hardware recommendations.</p><p>What you need to know:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QYfMfTr5dewsx1PWliVDVg.png" /><figcaption>Figure 1. Four critical hardware requirements</figcaption></figure><ul><li><strong>Your edge devices need to run Linux.</strong> Most industrial gateways, embedded systems, and edge servers run some flavor of Linux. That’s your starting point. Though some platforms do support Windows, the majority of edge deployments are Linux-based.</li><li><strong>Compute requirements scale with your models.</strong> Simple models run on modest hardware. Complex computer vision models need more horsepower.</li><li><strong>Connectivity matters more than you think.</strong> Edge deployment means your devices work when internet connectivity drops, but they still need to sync with your central platform when connectivity returns. Think about whether you need cellular backup, whether WiFi is reliable enough, or whether you need something more robust.</li><li><strong>Power and environment can’t be afterthoughts.</strong> Is this device running 24/7? Is it in extreme temperatures? Does it need to survive vibration, dust, or moisture? These constraints narrow your hardware choices fast.</li></ul><p>Once you’ve identified appropriate hardware, or at least narrowed down the category, the next piece is the intelligence itself.</p><h3>ML Models: The Intelligence</h3><p>Models need to actually work on edge devices, not just on your data scientist’s laptop. A model that performs beautifully during development can be completely unsuitable for edge deployment.</p><ul><li><strong>Size matters on edge devices.</strong> That 2GB model file? It won’t fit on a gateway with limited storage or it might max out available RAM.</li><li><strong>Inference speed is non-negotiable.</strong> If your use case needs real-time decisions, a model that takes 10 seconds to analyze each input won’t work.</li><li><strong>Format compatibility isn’t guaranteed.</strong> Models trained in different frameworks need to run on your edge infrastructure. Conversion and optimization are often necessary steps between training and deployment.</li><li><strong>Production data is messier than training data.</strong> Your model was trained on clean, well-lit images, but on a factory floor cameras get dusty, lighting varies throughout the day, and products occasionally appear at odd angles. Models need to be robust enough to handle real-world variation.</li></ul><p>The practical question is: how do you get from trained models to deployment-ready models?</p><p>Test on actual edge hardware early, or you may discover it’s too slow or too large for your target devices. Model optimization must become a standard step.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FUFLuah1VDRQSEv8VO7rQg.png" /><figcaption>Figure 2. ML model optimization pipeline</figcaption></figure><p>Your first deployed model won’t be your last. Performance drifts over time, new edge cases emerge, requirements change. You need infrastructure to retrain models with new data and push updates to deployed devices.</p><h3>ML Deployment Tooling: Getting Models to Devices</h3><p>Once you have trained models optimized for edge deployment, you need to actually get them onto your devices and keep them updated. This is where ML deployment tooling comes in.</p><p>Common deployment approaches:</p><p><strong>Container-based deployment:</strong> Package your model with dependencies into a Docker container. Deploy to edge devices. Update by pushing new versions. Works well if your team already uses containers and your edge hardware supports them.</p><p><strong>Configuration management:</strong> Tools like Ansible, Chef, or Puppet deploy model files, install dependencies, and manage updates. Good fit if you’re already using these for infrastructure management.</p><p><strong>Custom scripts:</strong> SSH into devices, copy model files, restart services. Works for small deployments. Doesn’t scale well, but sometimes simple is sufficient.</p><p><strong>ML-specific platforms:</strong> MLflow, Kubeflow, or vendor solutions provide purpose-built model deployment and versioning. Useful for sophisticated needs. Potential overkill for simple deployments.</p><p>The right choice depends on your team’s expertise and deployment scale. What matters is having a reliable way to deploy models, update them when you retrain, roll back if needed, and verify they’re running correctly.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/352/1*zzvUXHCjZb-tRT5ZgJqUAA.png" /><figcaption>Figure 3. Continuous deployment cycle for edge ML: Deploy models to devices, verify they’re running correctly, update when you retrain with new data, and maintain rollback capability when new versions underperform.</figcaption></figure><p><strong>Example:</strong> Your data science team exports a trained model. DevOps packages it into a Docker container. They push it to edge devices using standard tools. When you retrain next month, you deploy a new container version the same way. If it has issues, you roll back.</p><p>Once models are deployed and running locally, the next challenge is orchestrating everything around them — that’s where your IoT platform comes in.</p><h3>IoT &amp; Edge Platform: The Orchestration Layer</h3><p>Let me describe what happens when companies try to deploy edge ML without proper IoT platform infrastructure.</p><p>They start simple — one device, one model. Engineer SSHes into the device, sets up the ML inference endpoint, configures it to read from sensors. Works great.</p><p>Then they scale to ten devices. Now they’re managing ten separate edge systems. A bit tedious, but manageable with spreadsheets and scripts.</p><p>Next come fifty devices across multiple sites to manage. Any change to data pipelines, alert rules, or configurations becomes a week-long project. When devices go offline, nobody knows until someone visits the site. When data pipelines break, there’s no centralized detection.</p><p>Six months later, they have ML models that work fine locally, but the infrastructure around them is a mess. They’re spending more time maintaining IoT plumbing than improving ML models.</p><p>This is where IoT orchestration platforms matter — not to deploy your ML models, but to manage everything around them.</p><p>What you actually need:</p><ul><li><strong>Centralized device management.</strong> You need to see all your edge devices in one place. When you have fifty devices at remote sites, manual tracking isn’t realistic.</li><li><strong>Data pipeline orchestration.</strong> Sensor data needs to flow to your ML inference endpoints, then to dashboards, alerts, and downstream systems. This data routing shouldn’t require custom scripts on every device.</li><li><strong>Rule engine at the edge.</strong> The prediction logic needs to run locally (even when connectivity drops) and be manageable centrally.</li><li><strong>Handle offline operation gracefully.</strong> Edge devices keep running when connectivity drops. That means local data buffering, continuing to process sensor data, running your ML inference, executing alert rules. When connectivity returns, sync results with your central platform. Your ML models don’t care about network status, but the infrastructure should be able to handle it.</li><li><strong>Monitor the whole system centrally.</strong> You need visibility across all devices and sites. Even if your ML models are working perfectly on each device, without central monitoring, you’re flying blind.</li></ul><p>While you could build all this yourself, you may soon realize that you’re spending more time on IoT infrastructure than ML models, maintaining a platform instead of improving predictions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/543/1*tMxJThBEdWy7_VPRIamz8g.png" /><figcaption>Figure 4. Edge platform deployment: Central platform handles device management, configuration updates, security, and data aggregation across all sites. Local edge instances at each factory enable offline operation and sync when connectivity is available.</figcaption></figure><p>This is exactly what IoT platforms like ThingsBoard Edge handle. It’s purpose-built for edge orchestration. Your ML inference endpoints run however you want them to. ThingsBoard manages the IoT infrastructure around them — collecting sensor data, routing it to your models, processing results, triggering actions, syncing with central systems.</p><h3>Putting It Together: What Implementation Actually Looks Like</h3><p>Let’s make this concrete with a realistic scenario.</p><p>Imagine a manufacturing company with production equipment across multiple facilities. They’re doing predictive maintenance by detecting bearing failures before they cause downtime. This matches one of the use cases from our decision framework: high-value assets where downtime is expensive and you need advance warning.</p><p>Here’s what their implementation looks like:</p><ul><li><strong>Hardware layer:</strong> They have industrial gateways at each production line. These gateways are already there collecting vibration data, temperature readings, and other sensor telemetry. Each gateway has enough compute to run inference locally.</li><li><strong>Model layer:</strong> Trained models predict bearing failures based on vibration patterns. They know what healthy bearings look like versus bearings showing early signs of wear. Also, models are optimized to run on the gateway hardware — small enough to fit in memory, fast enough to analyze data every few minutes.</li><li><strong>Deployment &amp; Platform layers: </strong>ML models are deployed to each gateway as Docker containers using their existing DevOps pipeline. ThingsBoard Edge runs on each gateway, orchestrating the data flows. Each gateway runs inference locally. When ML models forecast increasing vibration levels indicating bearing degradation, the rule engine flags it and triggers alerts. Maintenance teams can see both historical patterns and future predictions to schedule repairs during planned downtime. All predictions and system status flow back to the central platform for monitoring.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/606/1*H7iuYFo62A4UWRmxobBYjA.png" /><figcaption>Figure 5. Edge ML architecture for predictive maintenance: ML models run in containers on industrial gateways, while ThingsBoard Edge orchestrates data flows, rule execution, and sync with the central platform. The dashboard shows real-time vibration data and ML-predicted bearing degradation.</figcaption></figure><p>When maintenance teams see a bearing flagged for degradation, they schedule replacement during planned downtime. No more catastrophic failures in the middle of production runs. No more emergency repairs.</p><p>Here’s what makes this work:</p><ul><li><strong>Devices operate independently.</strong> If a factory loses internet connectivity, the gateways keep running inference on local data. When connectivity returns, they sync results with the central platform. Production doesn’t stop just because the internet is down.</li><li><strong>Model deployment is decoupled from IoT orchestration.</strong> When they need to update ML models, they push new Docker containers through their standard deployment pipeline. When they need to adjust data flows, alert thresholds, or monitoring rules, they configure that through ThingsBoard. These are separate concerns handled by appropriate tools. They’re not trying to make one platform do everything.</li><li><strong>Monitoring is comprehensive.</strong> The maintenance team sees predictions from all sites in one dashboard. They can spot patterns. For example, they may notice that one facility’s bearings degrade faster due to environmental conditions or that certain equipment types have different failure signatures.</li></ul><p>This setup scales. They started with five production lines. Now it’s thirty lines across multiple facilities. Adding new sites means deploying another gateway and connecting it to the platform. The infrastructure doesn’t get more complex as they grow.</p><h3>Common Mistakes to Avoid</h3><p>I’ve seen enough edge ML deployments to know where teams typically stumble. Save yourself some pain:</p><h4><strong>Underestimating edge hardware constraints.</strong></h4><p>Your model works great on your laptop but takes 30 seconds on an edge device with limited RAM. Test on actual target hardware early and optimize models for edge constraints before deployment.</p><h4><strong>Building custom infrastructure instead of using platforms.</strong></h4><p>This is the big one. Smart teams think “we can build this ourselves.” And they can. But they end up spending six months building deployment automation, monitoring, device management, security — all the infrastructure that platforms already solved. Meanwhile their actual ML use case isn’t progressing. Use existing infrastructure where it makes sense. Build custom only where you genuinely need something unique.</p><h4><strong>Skipping offline operation testing.</strong></h4><p>It’s easy to build a system that works perfectly with reliable connectivity. Then you deploy to actual sites with spotty internet, and everything falls apart. Test what happens when devices lose connectivity.</p><h4><strong>No rollback plan for model updates.</strong></h4><p>Always have rollback capability for situations where a new model version performs worse than the old one.Common Mistakes to Avoid</p><h4><strong>Treating edge devices as trusted.</strong></h4><p>Edge devices are physically accessible. Someone could walk up to them, plug in a keyboard, and potentially compromise them. Especially in remote installations or publicly accessible locations. Security can’t be an afterthought. Encrypted communication, proper authentication, secure credential storage need to be built in from the start.</p><h4><strong>Conflating IoT orchestration with ML deployment.</strong></h4><p>You need tooling to deploy and update your models (Docker, Ansible, MLflow, custom scripts, or whatever fits your workflow). You also need an IoT platform to orchestrate data flows, execute rules, manage devices, and monitor the entire system. These are separate layers with different responsibilities. Don’t expect your IoT platform to serve as your ML deployment platform, nor vice versa.</p><h3>Where to Go From Here</h3><p>You’ve validated your use case in our first article. You’ve confirmed you have quality data. Now you understand what you need to actually implement: appropriate edge hardware, trained ML models, and a platform to orchestrate everything.</p><p>The smart approach is starting with proven infrastructure instead of building from scratch. Let someone else handle the deployment mechanics, device management, and monitoring complexity. Focus your effort on what’s unique about your use case — your specific equipment, your specific models, your specific predictions.</p><p><strong>Ready to implement edge orchestration?</strong> ThingsBoard Edge handles device management, data routing, rule execution, and monitoring across distributed deployments — while you control your ML layer.<a href="https://thingsboard.io/docs/edge/?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=ml_p2_edge_ml_deployment"> Explore the platform</a> or<a href="https://thingsboard.io/docs/user-guide/install/edge/installation-options/?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=ml_p2_edge_ml_deployment"> start with the free community edition</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c803e62695c4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Machine Learning in IoT: When It’s Worth the Effort, and When It’s Not]]></title>
            <link>https://medium.com/@ymala/machine-learning-in-iot-when-its-worth-the-effort-and-when-it-s-not-0e9a1c2286c8?source=rss-61c1ec8ecc92------2</link>
            <guid isPermaLink="false">https://medium.com/p/0e9a1c2286c8</guid>
            <category><![CDATA[iot]]></category>
            <category><![CDATA[use-cases]]></category>
            <category><![CDATA[edge-computing]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Yevheniia Mala]]></dc:creator>
            <pubDate>Wed, 26 Nov 2025 15:02:28 GMT</pubDate>
            <atom:updated>2026-04-15T09:02:56.477Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*89-EsEjPEjIJfyY2dpvdRg.png" /></figure><h3>ML in IoT: When to Use It and When to Skip It</h3><p>Most IoT projects succeed on data visibility, not data science. I’ve seen companies add ML to systems that already work perfectly well with simple rules — and I’ve seen cases where ML transforms a project from functional to exceptional. Most IoT systems don’t need ML, and the ones that do often fail because they add unrequired complexity too early. Here’s a framework for telling the difference.</p><h3>The Power of Rule-Based Systems</h3><p>Most IoT platforms give you analytics right out of the box. Platforms like <a href="https://thingsboard.io/?fpr=ai432j&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=ml_p1_ml_vs_rule-based">ThingsBoard</a> give you reports, dashboards, KPIs, and trend tracking. Rule engines let you automate complex processes with straightforward `if-then` logic. Systems like SCADA can run entire industrial operations from one control room. These conventional tools handle the majority of IoT work. For a lot of companies, that’s all they’ll ever need.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AywlKdEhwPuMYWQOWUTeTw.png" /></figure><p>Here are two examples:</p><p><strong>Temperature Monitoring in Cold Storage</strong></p><ul><li><strong>Simple solution: </strong>Alert triggers when temperature goes above 5°C</li><li><strong>Machine learning solution:</strong> Train a model to predict temperature drift patterns</li><li><strong>Reality check:</strong> For most facilities, a basic threshold catches problems fast enough. Drift prediction adds complexity without meaningful lead time.</li></ul><p><strong>Inventory Level Management</strong></p><ul><li><strong>Simple solution</strong>: Reorder when stock falls below predetermined threshold</li><li><strong>Machine learning solution</strong>: Demand forecasting with seasonal pattern recognition</li><li><strong>Reality check</strong>: Unless you’re dealing with highly volatile demand patterns, simple thresholds work fine.</li></ul><p>Rule-based systems are predictable, debuggable, and transparent. When a temperature alert fires, you know exactly why — because the value crossed your threshold. This simplicity is a feature, not a limitation.</p><h3>When Machine Learning Delivers Value</h3><p>So, when should you actually consider ML? The answer is simple: when a project is too complex to be managed by simple `if-then` rules.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/528/1*EmC5tmf3fhBbrdbz0Jot_g.png" /></figure><p>Here are the scenarios where the complexity pays off:</p><ul><li><strong>Pattern Recognition in Complex, Multi-Variable Systems</strong></li></ul><p><strong>The Use Case</strong>: Manufacturing quality control where dozens of variables interact in non-obvious ways.</p><p>ML models can spot correlations across multiple sensor inputs that you’d never define manually. The system learns which combinations predict defects — catching quality issues that even experienced operators might miss. This can cut defect rates significantly while keeping false alarms low.</p><ul><li><strong>Predictive Maintenance for High-Value Assets</strong></li></ul><p><strong>The Use Case</strong>: Predicting equipment failures in scenarios where downtime costs a fortune.</p><p>This is probably the most widespread application of ML in IoT. Models pick up on subtle shifts in vibration patterns or thermal signatures that signal bearing wear or mechanical stress — sometimes weeks before anything breaks. You get to schedule maintenance during planned downtime instead of dealing with catastrophic failures and emergency repairs.</p><ul><li><strong>Anomaly Detection in Rapidly Changing Environments</strong></li></ul><p><strong>The Use Case: </strong>Anomaly detection in utility and energy systems great for smart meters, HVAC, water systems, energy monitoring, etc.</p><p>ML models learn the patterns of energy consumption in different situations. Energy usage changes due to occupancy, weather, time of day, equipment cycles, or seasonal factors, and it’s not possible to describe every situation with basic rules. ML models identify deviations in consumption patterns and signal about equipment malfunction, leaks, stuck valves, failing HVAC components, and abnormal load patterns in each situation.</p><h3>Getting Started: Implementing Decisions</h3><p>If your use case matches one of these scenarios and you’ve decided machine learning can bring value to your IoT project, you’ll need to make decisions about your data, your implementation approach, and where to run the models.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/785/1*78tNVOpEPSBUu5DcZkZPag.png" /></figure><h4>The Hidden Barrier: Data Quality</h4><p>Here’s the part most people overlook: your ML model will only be as good as your data. The numbers back this up: <a href="https://www.iotforall.com/the-importance-of-data-quality-for-successful-ai">66%</a> of organizations cite poor data quality as a major obstacle to ML adoption. <strong>Even minor issues in your training data can cascade into major model failures.</strong></p><p>So, what exactly does <strong>“high-quality data”</strong> mean?</p><ul><li><strong>Accuracy: </strong>The data correctly reflects the real-world events.</li><li><strong>Completeness: </strong>There are no missing values or gaps in the data.</li><li><strong>Consistency: </strong>Data is uniform across all datasets and systems.</li><li><strong>Timeliness: </strong>The data is recent enough to be relevant.</li><li><strong>Relevance: </strong>The data directly applies to the problem you are trying to solve.</li></ul><p>With data quality handled, you’re ready to choose your implementation path.</p><h3>3 Approaches to Implementing Machine Learning in IoT</h3><p>You have three options, each with different trade-offs in terms of effort, customization, and expertise required.</p><h4>Platform-Native Machine Learning</h4><p>If your IoT platform has built-in machine learning capabilities, start here. Such tools let you build predictions and forecasts without writing training code. You configure models through the platform interface, and they work natively with your existing data structure.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/565/1*ofcFdH8a73tVcrZG_2ww8A.png" /></figure><p>For most IoT projects, platform-native machine learning is the smartest starting point. It’s already integrated, doesn’t require infrastructure setup, and you can start experimenting immediately.</p><h4>Pre-trained Model from Third-Party Vendors</h4><p>If platform-native capabilities don’t fit your needs but building custom models isn’t justified, pre-trained models are another option. They deploy quickly, require no programming expertise, and include vendor support. However, they come with ongoing licensing costs, are limited to the vendor’s design choices, and still require integration work with your IoT infrastructure.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/565/1*M2122HE_gmcjJwNt9WZEWw.png" /></figure><h4>Build a Custom Model</h4><p>Building a model from scratch gives you full control over training and customization. However, it also requires ML expertise, time, and investment in training infrastructure, such as purchasing GPU hardware or renting cloud GPU instances. Additionally, you need high-quality datasets with thousands to millions of labeled examples.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/565/1*uX5lz4yI81mtdvGLj8o5og.png" /></figure><p>Once you’ve chosen your implementation approach, the next decision is where to run your models.</p><h3>ML Model Deployment: Cloud vs. Edge</h3><p>Where you run your ML models has less to do with which approach you picked and more with what your project actually needs — how reliable your connectivity is, how fast responses need to be, and what infrastructure you already have in place.</p><p><strong>Cloud deployment </strong>runs your ML model as a web service. Devices send data to your IoT platform, which processes it through ML models — either built-in or via external API. This adds a network round-trip to every prediction. This works when:</p><ul><li>You have reliable, high-bandwidth,<strong> </strong>low-latency connectivity</li><li>Real-time response isn’t critical (you can tolerate 100–500ms latency)</li><li>You’re okay with data leaving your premises</li><li>You want someone else managing the infrastructure</li></ul><p><strong>Edge deployment</strong> runs models locally on devices or gateways at each site. Data stays local, inference is nearly instantaneous, and systems work even when internet connectivity drops. This matters when:</p><ul><li>Latency is critical (manufacturing lines, autonomous systems, safety applications)</li><li>Connectivity is unreliable (remote sites, mobile equipment)</li><li>Data privacy or sovereignty requires local processing</li><li>You’re paying per-message fees to cloud services at scale</li></ul><p>Understanding your options is one thing. Deciding whether ML is worth it for your specific project is another.</p><h4>Before you commit to machine learning, ask:</h4><ul><li>Does a rule-based solution already solve the problem?</li><li>Is the environment too complex for simple logic?</li><li>Do I have reliable, high-quality data?</li><li>Can I maintain that data quality over time?</li><li>Do I understand where the model will run (cloud or edge)?</li><li>Is the expected benefit greater than the cost and complexity?</li></ul><p>If you can’t answer “yes” to most of these, ML may not be worth it — yet.</p><h3>Conclusion: A Strategic Choice</h3><p>Deciding to use machine learning in an IoT project comes down to strategy, not technology. Before jumping in, make sure the problem actually needs ML instead of simpler rules. Check that you can get quality data and keep it that way — not just grab it once and hope. And figure out whether cloud or edge deployment makes sense for your connectivity and latency situation.</p><p>The best IoT systems aren’t the fanciest ones — they’re the ones where complexity matches the problem. Start with simple solutions, bring in ML when you’ve got a solid reason for it, and nail down your basics before adding layers</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0e9a1c2286c8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Brazil Spends Billions on Waste Management — Yet 41% Still Goes to Illegal Dumps]]></title>
            <link>https://medium.com/@ymala/why-brazil-spends-billions-on-waste-management-yet-41-still-goes-to-illegal-dumps-a19a53b6bf26?source=rss-61c1ec8ecc92------2</link>
            <guid isPermaLink="false">https://medium.com/p/a19a53b6bf26</guid>
            <category><![CDATA[smart-waste-management]]></category>
            <category><![CDATA[waste-management]]></category>
            <category><![CDATA[edge-computing]]></category>
            <category><![CDATA[brasil]]></category>
            <category><![CDATA[iot]]></category>
            <dc:creator><![CDATA[Yevheniia Mala]]></dc:creator>
            <pubDate>Tue, 18 Nov 2025 12:01:17 GMT</pubDate>
            <atom:updated>2025-11-18T12:01:17.841Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*InrghcJtn8RWfoN_fKjaUw.png" /></figure><h3>Why Brazil Spends Billions on Waste Management — Yet 41% Still Goes to Illegal Dumps</h3><p>In Brazil, waste collection runs locally. All 5,571 municipalities handle their own pickup, transportation, and final disposal. Most operate on fixed schedules and predetermined routes, regardless of actual demand.</p><p>Although this approach is practical, it ignores reality. Population density varies. Commercial districts fill bins faster than residential ones. But the trucks keep following the same routes, burning fuel to collect half-empty bins in one neighborhood while bins overflow onto sidewalks three blocks away. Brazil spends billions on waste management. However, <a href="https://orizonvr.com.br/en/main-challenges-of-solid-waste-management-in-brazil/">41% of the country’s 81.8 million tons of annual waste</a> still ends up in illegal dumps.</p><p>This article is the fourth part of our series on IoT in Brazil. In earlier stories, we explored <a href="https://medium.com/@ymala/one-country-two-internets-how-edge-computing-solves-brazils-connectivity-crisis-461835a98882">Brazil’s connectivity paradox</a>, <a href="https://medium.com/@ymala/lifesaving-care-how-edge-computing-is-changing-assisted-living-789ad10ae3b7">life-saving healthcare monitoring in remote areas</a>, and <a href="https://medium.com/@ymala/how-edge-computing-protects-brazils-remote-coffee-farms-71c1bb5398a0">crop protection on coffee farms</a>. Now, we turn to urban waste management and try to understand why billions in spending hasn’t solved illegal dumping, and why the right technology choices matter more than ever.</p><h3>Why Illegal Dumping Persists</h3><p>As mentioned above, local municipalities are responsible for their own waste collection system. About 70% of municipalities have fewer than 20,000 inhabitants, and their budgets cannot cover proper waste disposal. Recent studies show that municipalities would need to <strong>double</strong> their waste management budgets, <a href="https://www.researchgate.net/publication/383177383_Solid_Waste_Management_in_Small_Municipalities_When_and_How_to_Discuss_a_New_Fee">from 3.64% to 7.48%</a> of annual spending, just to meet legal compliance.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yjVKusoKusJKi0FvA4uZjg.png" /><figcaption>Table 1. Required vs. Actual Spending for Waste Management Compliance</figcaption></figure><p>That puts officials in an impossible position: pay employees or build compliant landfills? The choice usually goes to payroll. As a result, waste flows to the cheapest disposal option, legal or not. That’s why <a href="https://orizonvr.com.br/en/main-challenges-of-solid-waste-management-in-brazil/">over 2,000 open-air dumps and irregular landfills were still operating</a> in 2024.</p><p>Geography makes things even worse. Brazil is massive, and even when laws exist on paper, there aren’t enough inspectors to check thousands of disposal sites scattered across the country. Without real-time tracking of where waste actually goes after trucks pick it up, illegal dumping just keeps happening.</p><h3>The Cost of Inefficiency</h3><p>Brazil’s waste management problem isn’t just environmental — it’s expensive. The country lost <a href="https://agenciabrasil.ebc.com.br/geral/noticia/2022-06/indice-de-reciclagem-no-brasil-e-de-4-diz-abrelpe">R$14 billion</a> ($3.54 billion) in 2019 by landfilling materials that could have been recycled. That’s money thrown away along with the waste.</p><p>In 2023, the national recycling rate was only about <a href="https://www.abrema.org.br/2024/12/12/reciclagem-de-residuos-chega-a-8-no-pais-com-trabalho-informal-aponta-estudo/">8%</a>, which is half the global average of 16%. The National Solid Waste Plan <a href="https://www.trade.gov/market-intelligence/brazil-waste-management">targets 48% recycling rate by 2040, which requires $6 billion in new infrastructure investment over 15 years.</a> Meanwhile, municipalities spent <a href="https://orizonvr.com.br/en/main-challenges-of-solid-waste-management-in-brazil/">R$ 37 billion</a> ($7.4 billion) just keeping the current system running<strong> </strong>in 2023.</p><p>The current approach cannot deliver results. When collection trucks drive inefficient routes, when recyclables end up in dumps, when there’s no data to guide decisions, every real spent delivers diminishing returns. The system needs visibility into what’s actually happening.</p><h3>Smart Waste Management​ Addresses the Problem</h3><p>How do you track thousands of waste bins across a city? Most Brazilian municipalities don’t. They have no idea which containers are full, and which are empty. Smart cities use IoT sensors to detect exactly this. To get similar results, municipalities need sensor networks that track bin fill levels and transmit data for route optimization. This shift from fixed schedules to demand-based collection can significantly cut operational costs.</p><p>The technical setup is straightforward. Sensors installed in waste bins measure fill levels and transmit data wirelessly — typically via LoRaWAN for wide coverage or cellular networks where infrastructure exists. This data flows to either cloud platforms or local edge devices, where it gets processed in real time. For Brazil’s context, edge computing matters. When internet connectivity drops in smaller municipalities, local processing keeps the system running. Data syncs automatically when connections return.</p><p>What changes operationally? Instead of trucks following fixed Monday-Wednesday-Friday routes, collection gets triggered by actual need. A dashboard shows operators which bins are above 80% capacity. Route optimization algorithms calculate the most efficient path to service only those bins.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4isljh6PdCcPZxmbgTrRkw.gif" /><figcaption>Figure 1. Smart waste management dashboard</figcaption></figure><p><a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC11290616/">Studies</a> from pilot deployments show measurable results versus conventional practices:</p><ul><li>32% improvement in route efficiency</li><li>29% decrease in fuel consumption and emissions</li><li>33% increase in waste processing throughput</li><li>18% vehicle maintenance savings</li></ul><p>For municipalities facing the budget constraints described earlier, these changes can make proper waste management affordable.</p><p><strong>See it in action:</strong> Interactive <a href="https://thingsboard.io/use-cases/waste-management/?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_p4_waste_management">waste management demo</a> shows how edge computing handles offline operation.</p><h3>IoT Platform Comparison</h3><p>The efficiency gains depend entirely on platform choice, especially in Brazil with its connectivity challenges. Not all IoT solutions deliver equally. Understanding the differences between their architectures will protect against wasted investment.</p><p>Before starting a project and choosing a platform, decide on the purpose of the deployment. Answering these questions (and others like them) clarifies which platform fits best:</p><ol><li><strong>What problem are we solving?</strong> Reduce fuel costs? Improve recycling rates? Cut overflow response times?</li><li><strong>What infrastructure do we already have?</strong> Digital maps? Fleet tracking? Asset registries?</li><li><strong>How stable is our connectivity?</strong> Does the internet stay up, or do we face frequent outages?</li><li><strong>What is our budget?</strong> Can we deploy city-wide, or should we start with pilot zones?</li></ol><p>The market offers different approaches, such as: managed cloud services, open-source platforms, commercial editions with enterprise features and support. Let’s delve into the existing options, each of which has its own set of tradeoffs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1cpcZFgABejqZn_4vvnxuA.png" /><figcaption>Table 1. IoT Platform Comparison</figcaption></figure><p>For context: estimated costs based on typical deployments of 1,000 smart bins run roughly $15,000-$25,000 annually with managed cloud services, $13,000-$20,000 with commercial open-source editions, or $5,000-$10,000 (plus internal staff time) with self-hosted open-source.</p><p>The choice depends on a combination of what you already have and what you can afford. Want to keep the costs low? Go for open-source options. Facing frequent connectivity issues? Platforms with native edge support keep operations running during outages. Need to deploy fast? Commercial editions split the difference. Already running your operations on AWS or Azure? Managed cloud services fit naturally into that ecosystem.</p><h3>Seeing the Bigger Picture</h3><p>Brazil doesn’t need to spend more on waste management — it needs to spend smarter. The technology already exists to shed light on what’s been hidden for decades: overflowing bins, idle trucks, recyclables lost before they’re ever reused. The difference lies in how cities collect, process, and act on data.</p><p>IoT and edge computing give municipalities the power to see their operations clearly — even when the internet goes down. But visibility isn’t the finish line; it’s the start of accountability, efficiency, and better public service. When data becomes part of daily decision-making, every dollar spent can finally deliver real results.</p><p>This isn’t just about waste. From remote healthcare centers to coffee farms and now urban infrastructure, Brazil’s IoT journey tells one consistent story — that progress starts when information moves closer to where action happens.</p><p>In a country this vast and complex, the right platform doesn’t just connect devices; it connects people, places, and priorities. When local actions align with national goals, chaos becomes coordination — and technology becomes a tool for everyday resilience.</p><p><strong>Deploy it yourself:</strong> <a href="https://github.com/thingsboard?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_p4_waste_management">ThingsBoard</a> open-source platform | <a href="https://github.com/thingsboard/thingsboard-edge?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_p4_waste_management">ThingsBoard Edge</a> for offline capability</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a19a53b6bf26" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How Edge Computing Protects Brazil’s Remote Coffee Farms]]></title>
            <link>https://medium.com/@ymala/how-edge-computing-protects-brazils-remote-coffee-farms-71c1bb5398a0?source=rss-61c1ec8ecc92------2</link>
            <guid isPermaLink="false">https://medium.com/p/71c1bb5398a0</guid>
            <category><![CDATA[brazil]]></category>
            <category><![CDATA[smart-irrigation]]></category>
            <category><![CDATA[edge-computing]]></category>
            <category><![CDATA[smart-farming]]></category>
            <category><![CDATA[iot]]></category>
            <dc:creator><![CDATA[Yevheniia Mala]]></dc:creator>
            <pubDate>Wed, 29 Oct 2025 16:17:19 GMT</pubDate>
            <atom:updated>2026-02-19T13:22:00.732Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cjeScKx0sw9Y4JZk78HeJA.png" /></figure><p>Droughts, unpredictable rains, and pests can wipe out as much as half of a harvest. For Brazil, which supplies 40% of global coffee, these aren’t hypothetical risks. Under extreme weather conditions, farms can lose an estimated <a href="https://www.spilling-the-beans.net/exclusive-reality-hit-brazil-coffee-growers-struggle-to-recover/">20% — 50% </a>of annual yields. Brazil’s coffee exports generated $10.37 billion in 2024 (Statista), making these losses economically devastating for both individual farms and the national economy.</p><p>Traditional farming relies on experience and scheduled interventions, but climate change and rising global demand have made this approach insufficient. On top of that, many plantations are located in remote regions, far from reliable infrastructure. Farmers face labor shortages precisely when they need more intensive monitoring and faster response time.</p><p><a href="https://thingsboard.io/docs/edge/getting-started-guides/what-is-edge/?fpr=ai432j&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part3_irrigation">Edge computing</a> changes this equation. By processing data locally and automating responses without depending on constant connectivity, platforms like <a href="https://thingsboard.io/products/thingsboard-edge/?fpr=ai432j&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part3_irrigation">ThingsBoard Edge</a> allow farmers to monitor soil conditions, control irrigation, and prevent crop loss — even when the internet fails for days. This is exactly why Brazil’s most progressive coffee farmers are adopting edge-enabled IoT solutions.</p><p>This article is the third part of our series on IoT in Brazil. In earlier stories, we explored <a href="https://medium.com/@ymala/one-country-two-internets-how-edge-computing-solves-brazils-connectivity-crisis-461835a98882?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part1">Brazil’s connectivity paradox</a>, and how edge computing enables <a href="https://medium.com/@ymala/lifesaving-care-how-edge-computing-is-changing-assisted-living-789ad10ae3b7">life-saving healthcare monitoring in remote communities</a>. Now, we explore how the same technology protects Brazil’s most valuable crop.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_OGwH-uK-YFh-zVqcLZ2tA.png" /></figure><h3>How Edge Computing Makes It Work</h3><p>Remote coffee farms need an IoT platform that continues to operate when the Internet fails, and syncs with cloud for analytics when connectivity returns. ThingsBoard’s Edge-Cloud architecture was designed precisely for this scenario. Here’s how the components work together to provide continuous monitoring and automation regardless of network conditions:</p><p><strong>Collect</strong></p><p>ThingsBoard Edge integrates with any sensor type via the variety of supported protocols. In coffee farming deployments, this typically includes:</p><ul><li><strong>Soil sensors:</strong> Moisture, pH, NPK nutrients, temperature (LoRaWAN/Modbus)</li><li><strong>Weather stations:</strong> Rainfall, humidity, wind, solar radiation (MQTT)</li><li><strong>Crop monitoring:</strong> Drone imagery, pest traps, disease detection cameras</li><li><strong>Equipment telemetry:</strong> Pump status, valve positions, water tank levels</li></ul><p>All sensors transmit to local gateways, which aggregate and normalize data in real-time — no cloud connection required.</p><p><strong>Connect</strong></p><p>For sensor communication, LoRaWAN provides 5–15km range with minimal power consumption — ideal for coffee plantations where cellular coverage is spotty. Sensors can operate for years on battery power while continuously feeding data to the local edge gateway.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GTqjH_WEheYdt58UdajalQ.png" /></figure><p><strong>Visualize</strong></p><p>ThingsBoard dashboards provide multiple visualization layers accessible from desktop or mobile:</p><ul><li><strong>Farm overview:</strong> Interactive map displaying all sensor locations, current readings, and active alerts</li><li><strong>Zone details:</strong> Drill down into specific plots to see soil moisture trends, irrigation history, and yield comparisons</li><li><strong>Equipment status:</strong> Monitor pump operations, valve positions, and system health in real-time</li><li><strong>Historical analysis:</strong> Compare conditions across seasons, identify patterns, optimize resource allocation</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dzDsOXZOSf4fNyGgx79iWQ.gif" /><figcaption><em>Figure 1. Smart irrigation dashboard</em></figcaption></figure><p>Farm managers access these dashboards via web browser, even during internet outages. See how <a href="https://thingsboard.io/iot-data-visualization/?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part3_irrigation">IoT data visualization</a> works in practice.</p><p><strong>Automate</strong></p><p>ThingsBoard’s Rule Chain engine transforms data into action via:</p><ul><li><strong>Complex multi-condition logic:</strong></li></ul><pre>// Automated Irrigation Rule Example<br>IF (soil_moisture &lt; 25% <br>    AND weather_forecast.rain == false <br>    AND water_reservoir &gt; 40%) <br>THEN {<br>    activate_drip_irrigation(duration: 2hrs, zones: affected_zones);<br>    send_notification(manager, &quot;Irrigation activated in Zone A&quot;);<br>}</pre><ul><li><strong>Alert prioritization:</strong></li></ul><p>Critical events (pest detection spike) → SMS + push notification</p><p>Warnings (gradual pH drift) → Email summary</p><p>Info (scheduled maintenance due) → Dashboard notification</p><p>These rules are executed locally on Edge, all without internet dependency. Explore our <a href="https://thingsboard.io/use-cases/smart-irrigation/?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part3_irrigation">smart irrigation solution</a> showing how to manage farming processes intelligently.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MeqfQaAqas5WtOje1RN9mg.png" /><figcaption><em>Figure 2. ThingsBoard Edge architecture for coffee farm deployment. Sensors communicate via LoRaWAN to the local gateway, which processes all data and automation locally. ThingsBoard Cloud provides long-term analytics and remote access when connectivity is available.</em></figcaption></figure><p>This architecture addresses the fundamental challenge of remote IoT: maintaining operational continuity when connectivity is unreliable. For Brazilian coffee farms hundreds of kilometers from infrastructure, this is the difference between 24/7 crop monitoring and systems that go dark during multi-day outages. Explore real-world IoT <a href="https://thingsboard.io/case-studies/?category=Smart-agriculture?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part3_irrigation">success stories</a> in agriculture.</p><h3>Two Farms, Two Approaches: Traditional vs. Edge-Enabled Farm</h3><p>To clearly understand the real impact of an edge-driven strategy, let’s compare two hypothetical 100-hectare coffee farms in interior Minas Gerais with identical conditions. The first one (Farm A) follows traditional farming methods. The other one (Farm B) has deployed ThingsBoard Edge on site. Here’s what the operational differences look like:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CvUJvC1SHpKFaGLubIIJew.png" /><figcaption><em>Table 1. Side-by-side operational comparison. Note: Figures and behaviors are illustrative, collected from published research on IoT and precision-agriculture systems. Actual results will vary by region, soil, and implementation.</em></figcaption></figure><h3>Annual Results Comparison</h3><p>After 365 days, the small, daily advantages of the Edge-enabled farm compound into a significant financial return.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AKg4BTlwKGRJL3GjqMNwRg.png" /><figcaption><em>Figure 4. Annual results comparison</em></figcaption></figure><p>The following metrics should also be considered:</p><ul><li><strong>Crop Loss Prevention:</strong> Early detection of drought stress and pest outbreaks protects an estimated 10–15% more yield compared to reactive traditional methods — directly increasing harvest revenue.</li><li><strong>Resource Optimization:</strong> Precision irrigation reduces water consumption by 15–30%, while soil monitoring enables targeted fertilizer application reducing overuse by 20–25%. Combined input cost savings: R$25,000–40,000 ($5,000–8,000 USD) annually.</li><li><strong>Labor Efficiency:</strong> Remote monitoring reallocates ~40% of field surveillance time to higher-value tasks like pruning and harvest planning — increasing productivity without additional hiring.</li><li><strong>Equipment Uptime:</strong> Predictive maintenance alerts and instant failure detection minimize emergency repairs, preventing costly irrigation disruptions during critical growth periods.</li></ul><p>For a 100-hectare farm, these improvements can amount to hundreds of thousands of dollars in preserved revenue and cost savings annually.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vHBbzOWSAUmO9WfLaE6_rA.png" /><figcaption><em>Table 2. Side-by-side annual costs comparison. Note: Figures and behaviors are illustrative, collected from published research on IoT and precision-agriculture systems. Actual results will vary by region, soil, and implementation.</em></figcaption></figure><p>By comparing these two approaches, we can estimate <strong>annual savings</strong> of up to<strong> $70,000.</strong></p><h3>The Path Ahead for Brazil’s Coffee Farms</h3><p>Brazil’s coffee industry won’t wait for perfect rural internet connectivity to arrive — and it doesn’t have to. While government initiatives continue expanding infrastructure and 5G networks gradually reach more regions, the farms need solutions that work today, in current conditions.</p><p>For Brazilian coffee farmers, this isn’t about adopting technology for technology’s sake. It’s about surviving droughts, responding to pest outbreaks in minutes instead of days, and optimizing resource use across plantations too vast for manual monitoring.</p><p>This is what we’re exploring in this series: whether protecting elderly patients in rural healthcare facilities, monitoring coffee harvests across remote plantations — the challenge remains consistent. Brazil’s digital transformation is happening in regions where connectivity can’t be guaranteed. Edge-first architecture isn’t a workaround; it’s the foundation.</p><p><strong>Next in this series:</strong> We’ll explore how Brazilian cities are using edge computing for smart waste management and urban resource optimization — solving logistics challenges in areas where traditional cloud platforms struggle with scale and connectivity.</p><p>Interested in exploring edge-powered smart farming? Try the <a href="https://thingsboard.io/products/thingsboard-edge/?fpr=ai432j&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part3_irrigation">open-source version</a> of the platform or <a href="https://thingsboard.io/docs/contact-us/?subject=ThingsBoard%20Products&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part3_irrigation">request a demo</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*K-Zs0DwrBhnvybhtf6zg4w.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=71c1bb5398a0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Lifesaving Care: How Edge Computing is Changing Assisted Living]]></title>
            <link>https://medium.com/@ymala/lifesaving-care-how-edge-computing-is-changing-assisted-living-789ad10ae3b7?source=rss-61c1ec8ecc92------2</link>
            <guid isPermaLink="false">https://medium.com/p/789ad10ae3b7</guid>
            <category><![CDATA[telehealth]]></category>
            <category><![CDATA[remote-patient-monitoring]]></category>
            <category><![CDATA[edge-computing]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[brazil]]></category>
            <dc:creator><![CDATA[Yevheniia Mala]]></dc:creator>
            <pubDate>Wed, 15 Oct 2025 13:58:06 GMT</pubDate>
            <atom:updated>2026-02-19T13:16:28.804Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tdRSX4d4GK31Rq3AfHNxyw.png" /></figure><p>Brazil has some of the fastest internet speeds in its cities and some of the weakest coverage in its rural regions. For healthcare facilities in those remote areas, this isn’t just an inconvenience, it’s a serious challenge. How do you monitor patients continuously when the internet can’t be trusted?</p><p>This is the question we explore in Part 2 of this series. Assisted living facilities in Brazil’s interior offer a clear example of what’s at stake — and how <a href="https://thingsboard.io/docs/edge/getting-started-guides/what-is-edge/?fpr=ai432j">edge computing</a> can make a difference.</p><h3>The Challenge of Continuous Care</h3><p>Demand for online healthcare has surged since the COVID-19 pandemic, especially for seniors and people with chronic illnesses. Healthcare facilities face the need to provide continuous patient monitoring, even when Internet connectivity is unreliable.</p><p>Traditionally, the only solution was to assign staff to check on patients around the clock. But even that isn’t sustainable. Caregivers can’t be everywhere at once, and the approach is both costly and exhausting.</p><p>Modern technology offers a smarter path. Wearable wristbands and smart sensors create a constant flow of data instead of occasional check-ins. Platforms like <a href="https://thingsboard.io/?fpr=ai432j&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part2_healthcare">ThingsBoard</a> compile this information and display it on real-time dashboards, giving caregivers a continuous view of each patient’s location and wellbeing — and the ability to respond immediately when something goes wrong. See how <a href="https://thingsboard.io/iot-data-visualization/?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part2_healthcare">IoT dashboards visualize data</a> in real time.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JoD7Yxd3jUdg6JDgIrjTWw.gif" /><figcaption><em>Figure 1. The ThingsBoard Edge dashboard showcases visualized data in an assisted living facility.</em></figcaption></figure><h3>Measuring the Impact</h3><p>Implementing IoT in healthcare facilities isn’t just a technical upgrade. While it requires upfront investment, the return on investment is compelling.</p><p>The benefits of implementing remote patient monitoring (RPM) systems are being studied around the world. While specific results vary by facility type, patient population, and implementation scope, these benchmarks demonstrate the potential impact:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s9I-8UHksH4w5JzmE1h7Fw.png" /><figcaption><em>Table 1: Relevant global benchmark numbers. Sources: </em><a href="https://electroiq.com/stats/remote-patient-monitoring-statistics/"><em>Remote Patient Monitoring Statistics and Facts</em></a><em> by S.Revankar; </em><a href="https://remingtonreport.com/remote-patient-monitoring-the-cost-savings-and-readmission-reductions-for-specific-medical-conditions-in-home-care/"><em>Remote Patient Monitoring: The Cost Savings and Readmission Reductions for Specific Medical Conditions in Home Care</em></a><em> by The Remington Report.</em></figcaption></figure><p>Of course, there are other metrics that demonstrate RPM efficiency, but mostly the results are achieved through alerting caregivers to potential emergencies before they turn into real ones.</p><h3>Why This Matters for Brazil</h3><p>Let’s consider Brazil’s rural context:</p><ul><li>Around <strong>16% of Brazilians</strong> (approximately 30 million people) <strong>live in rural areas</strong>, often far from hospitals and clinics.</li><li>Aging population, with growing chronic disease burden, particularly diabetes, hypertension, and heart disease.</li><li>Emergency medical transport often involves long-distance travel over unpaved roads.</li><li>Public and private health providers alike face rising costs.</li><li>Internet connectivity remains unreliable, with<strong> 42% of rural residents lacking network access</strong>.</li></ul><p>In such a context, the RPM systems become even more critical. Edge-enabled monitoring allows caregivers to catch problems early, prevent unnecessary hospital transfers, and reduce the burden of constant manual patient oversight. Solutions like <a href="https://thingsboard.io/products/thingsboard-edge/?fpr=ai432j&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part2_healthcare">ThingsBoard Edge</a> turn these capabilities into reality, helping providers deliver reliable care even where resources and connectivity are limited.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KQYv7jitBFVLwwd4UwVxVA.png" /></figure><h3>Inside the System: How Edge ComputingMakes It Work</h3><p>Both cloud and edge computing are capable of handling continuous health monitoring. However, considering patchy Internet coverage across Brazil, edge computing is crucial. With ThingsBoard Edge, all patient data is processed locally, so monitoring continues without interruption during Internet outages. Once connectivity returns, the system automatically synchronizes with ThingsBoard Cloud, ensuring nothing is lost.</p><p>Here’s how the technical implementation works:</p><ul><li>Wearable wristbands track vital signs like heart rate, oxygen levels, etc., while smart sensors monitor room conditions like temperature and humidity.</li><li>All data is sent through gateways via standard protocols such as BLE, LoRaWAN, or MQTT.</li><li>At the local level, ThingsBoard Edge processes the data in real time, linking it to specific residents, rooms, and zones.</li><li>If a resident’s vitals deviate from their baseline, the Rule Engine triggers automated alarms, sending instant notifications to caregivers.</li><li>Caregivers can access real-time dashboards that display alarms, patient status, and environmental conditions at any time.</li><li>Historical data can be sent to ThingsBoard Cloud for long-term storage, advanced analytics, and compliance reporting.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uWPiB2gATFhCA3o-3wBHBg.png" /><figcaption><em>Figure 2. ThingsBoard Edge architecture for living assistance deployment. Sensors communicate via LoRaWAN, BLE, or other protocols to the local gateway, which processes all data and automation locally. ThingsBoard Cloud provides long-term analytics and remote access when connectivity is available.</em></figcaption></figure><p>For healthcare providers, implementation speed often determines success. ThingsBoard Professional Edition (PE) offers pre-built healthcare monitoring templates — including dashboards, alert logic, and device profiles — enabling facilities to deploy pilot systems in days rather than weeks. The open-source Community Edition supports equivalent functionality with additional configuration time.</p><h3>Why It Pays Off</h3><p>Implementing RPM systems and IoT solutions like Thingsboard Edge isn’t only about the return on investments — it’s also about the direct benefit to the people you care for. Explore our <a href="https://thingsboard.io/use-cases/health-care/?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part2_healthcare">smart assisted living solution</a> showing how IoT-enabled monitoring works in healthcare. Here’s what that looks like in practice:</p><ul><li><strong>Prevent unnecessary hospital visits and readmissions.</strong> Continuous monitoring can detect early signs of trouble, allowing caregivers to step in before a crisis occurs.</li><li><strong>Optimize staff workload.</strong> Instead of spending hours on repetitive manual checks, caregivers can focus on residents who actually need attention.</li><li><strong>Lower infrastructure costs.</strong> Local data processing reduces reliance on constant, high-availability internet and keeps facilities from paying for expensive connectivity just to run safety systems.</li><li><strong>Shift from reactive to proactive care.</strong> Instead of waiting for the next check-in to uncover an issue, staff have a continuous, objective stream of health data for every resident.</li><li><strong>Improve wellbeing for everyone.</strong> Safer, calmer environments reduce burnout for staff while giving families peace of mind that their loved ones are always being looked after.</li></ul><h3>Beyond the Technologies</h3><p>Technology alone won’t solve the challenges of rural healthcare. What makes a difference is how it’s applied. In regions with near-absent connectivity, healthcare facilities can leap forward by combining simple devices with intelligent platforms like ThingsBoard. The result is a system that not only keeps patients safer but also helps staff work smarter, reduces costs, and strengthens the overall fabric of community care.</p><p>As connectivity continues to expand, these edge-first solutions will form the backbone of more resilient healthcare networks — not just in Brazil, but in remote communities everywhere.</p><p><strong>Next in this series:</strong> We’ll explore how Brazilian farmers are using IoT and edge computing for surviving droughts, responding to pest outbreaks, and optimizing resource use across plantations too vast for manual monitoring.</p><p>Interested in exploring edge-powered healthcare monitoring? Try the <a href="https://thingsboard.io/products/thingsboard-edge/?fpr=ai432j&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part2_healthcare">open-source version</a> of the platform or <a href="https://thingsboard.io/docs/contact-us/?subject=ThingsBoard%20Products&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part2_healthcare">request a demo</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=789ad10ae3b7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[One Country, Two Internets: How Edge Computing Solves Brazil’s Connectivity Crisis]]></title>
            <link>https://medium.com/@ymala/one-country-two-internets-how-edge-computing-solves-brazils-connectivity-crisis-461835a98882?source=rss-61c1ec8ecc92------2</link>
            <guid isPermaLink="false">https://medium.com/p/461835a98882</guid>
            <category><![CDATA[brazil]]></category>
            <category><![CDATA[internet-of-things]]></category>
            <category><![CDATA[edge-computing]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <dc:creator><![CDATA[Yevheniia Mala]]></dc:creator>
            <pubDate>Fri, 03 Oct 2025 13:27:43 GMT</pubDate>
            <atom:updated>2026-02-19T13:04:28.884Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*By_UidA1aKexPUKmuD8Pdw.png" /></figure><p>We don’t realize how accustomed we are to having easy access to the internet. Every morning, we sit in our cars and check for traffic or the latest news without giving it a second thought. We don’t think about how it works. It just works. But what if it didn’t?</p><p>Over the last decade, the Internet of Things (IoT) has changed the way we live and work. In countries like Brazil, the contrast is stark: a business owner in São Paulo enjoys 5G speeds while a farmer 200 kilometers away struggles with patchy 3G. This digital divide limits both business and personal opportunities. Despite the government initiatives, persistent obstacles continue to slow down the adoption of the IoT, ultimately holding back economic growth.</p><p><a href="https://thingsboard.io/docs/edge/getting-started-guides/what-is-edge/?fpr=ai432j&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part1">ThingsBoard Edge</a> offers a way forward. Unlike cloud-only IoT platforms that fail when connectivity drops, ThingsBoard Edge provides a complete local IoT infrastructure that processes data locally and automatically synchronizes with ThingsBoard Cloud when connections are restored.</p><p>In this series, we explore Brazil’s connectivity challenges and the technologies helping to overcome them, starting with the current state of Internet access.</p><h3>Connectivity in Brazil: A Mixed Reality</h3><p>5G Internet became available to residents of Brazil in 2022, and by the end of 2024, it ranked 39th in mobile Internet and 27th in fixed broadband according to <a href="https://www.speedtest.net/global-index/brazil">Ookla</a>. By August 2025, the mobile speed jumped from 80.97 Mbps to 239.43 Mbps, pushing Brazil up to the fourth place worldwide.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QgyHhihYhNOTfh-A_qylYA.png" /></figure><p>Despite the impressive rankings, reality tells a different story. While a commuter in Rio de Janeiro can stream real-time traffic updates on a 5G phone, a caregiver in Mato Grosso struggles to access the Internet. In fact, 2G and 3G devices remain in active use across rural Brazil simply because they are the only reliable option.</p><p>Coverage maps show the divide clearly:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ItApy8mb30dWMEl9qtDpUg.png" /></figure><p>For example, <a href="https://en.wikipedia.org/wiki/Claro_Company">Claro Mobile</a>, one of the largest telecom providers in Latin America, provides excellent 4G/5G service throughout São Paulo’s metropolitan area, yet the coverage in the State of Maranhão is much weaker.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UrsBUoGrpMqFq3O6K50xtQ.png" /></figure><p>The official statistics just confirm the gap. A June 2024 report by Anatel found that while 99.7% of urban Brazilians enjoy network coverage, 42.31% of rural residents have no network access at all.<strong> </strong>For context, that’s roughly 6 million people living in digital silence, including the farmers, manufacturers, and communities that form the backbone of Brazil’s economy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Xo2gEzZimKr-WQR7BiIexA.png" /></figure><p>For schools, clinics, and businesses in these regions, the implications are huge. Limited access means children missing out on online education, patients unable to use telemedicine, and small businesses stuck with outdated technology.</p><p>This is where edge computing proves invaluable. Instead of sending every piece of data to distant cloud servers (which requires constant internet), edge computing processes information right where it’s generated — in your factory, farm, or facility. Critical operations continue running regardless of connectivity issues, with all data automatically synchronized when internet service resumes.</p><p>This is exactly the approach <a href="https://thingsboard.io/products/thingsboard-edge/?fpr=ai432j&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part1">ThingsBoard Edge</a> takes: providing full IoT functionality locally while seamlessly integrating with ThingsBoard Cloud when connectivity allows.</p><h3>The Future of the IoT in Brazil</h3><p>The Brazilian government isn’t standing still. In 2019, it launched a National Plan for IoT, laying out strategies across education, innovation, tax regulation, security, and privacy.</p><p>Thanks to these initiatives, Brazil’s IoT market is experiencing constant growth. According to Grand View Research, Brazil’s cloud computing sector is projected to reach USD 41.73 billion by 2030 — a massive 23.8% annual growth rate.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*K7wqLNtVGaLrPjzogMdxNg.png" /></figure><p>Edge computing is growing twice as fast, with Grand View Research forecasting the edge market will reach USD 11.3 billion by 2030 at a 46.1% CAGR.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YkFZt5y9uzl0gsKS8DsXZA.png" /></figure><p>Government-backed initiatives are creating unprecedented demand for IoT solutions, but also highlighting the connectivity challenges that traditional cloud-only platforms can’t solve. This is precisely when edge computing becomes essential. This dual approach unlocks possibilities across Brazil’s key industries:</p><ul><li><strong>Agriculture:</strong> Automated irrigation systems that optimize water usage even when cellular towers are 50+ kilometers away</li><li><strong>Energy:</strong> Remote monitoring of wind farms and solar installations without waiting for spotty connections</li><li><strong>Logistics:</strong> Real-time fleet tracking through Brazil’s vast rural highways, ensuring no shipment data is ever lost</li><li><strong>Manufacturing:</strong> Smart factories that maintain full operational intelligence during network outages, preventing costly downtime.</li></ul><h3>Putting IoT into Practice</h3><p>While connectivity challenges persist across rural Brazil, the momentum for IoT innovation is unstoppable. With strong governmental support, faster mobile networks, and growing investment in both edge and cloud computing, the foundation for the next wave of digital transformation is already in place.</p><p>What does this look like in action?</p><p>In the second part of this series, we’ll take you inside real Brazilian deployments using ThingsBoard solution templates — from assisted living systems enhancing elderly care safety to precision agriculture systems optimizing irrigation in remote fields to smart city waste management streamlining collection routes. You’ll see implementation strategies, technical approaches, and proven methodologies that work today, even in Brazil’s most connectivity-challenged regions.</p><p>➡️ Stay tuned to see innovation at the heart of Brazil’s digital revolution — and discover how ThingsBoard Edge bridges the gap between unreliable connectivity and reliable IoT performance.</p><p><strong>📊 </strong><a href="https://thingsboard.io/iot-data-visualization/?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part1"><strong>Discover IoT data visualization↗</strong></a><strong><br>📚</strong><a href="https://thingsboard.io/case-studies/?fpr=ai432j?utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part1"><strong> Explore customer stories↗</strong></a><a href="https://thingsboard.io/case-studies/"><strong><br></strong></a><strong>🚀</strong><a href="https://thingsboard.io/products/thingsboard-edge/?fpr=ai432j&amp;utm_source=medium&amp;utm_medium=social&amp;utm_campaign=brazil_part1"><strong>Try ThingsBoard Edge today↗</strong></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=461835a98882" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>