<?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 Finlay Piroth on Medium]]></title>
        <description><![CDATA[Stories by Finlay Piroth on Medium]]></description>
        <link>https://medium.com/@finpiroth?source=rss-5d89cb499422------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*wJPtZJj0K1DCD5V_24NiQg.png</url>
            <title>Stories by Finlay Piroth on Medium</title>
            <link>https://medium.com/@finpiroth?source=rss-5d89cb499422------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 16:51:08 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@finpiroth/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[From Prompts to Agents]]></title>
            <link>https://generativeai.pub/from-prompts-to-agents-b6ddbfa407e8?source=rss-5d89cb499422------2</link>
            <guid isPermaLink="false">https://medium.com/p/b6ddbfa407e8</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[chatgpt]]></category>
            <dc:creator><![CDATA[Finlay Piroth]]></dc:creator>
            <pubDate>Tue, 22 Jul 2025 04:40:26 GMT</pubDate>
            <atom:updated>2025-07-22T04:40:26.337Z</atom:updated>
            <content:encoded><![CDATA[<h4>Mapping the Gen AI Maturity Curve</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*T6qXEdLjII46XEWD" /></figure><p>Just over two and a half years ago, the world changed forever when OpenAI publicly released the first version of ChatGPT. Since then, the pace of innovation has been staggering, as competitors quickly rose up with their own Large Language Models (LLMs). Over time, new systems, techniques, and technology have taken the spotlight, and models have continued to improve exponentially in nearly every measurable way.</p><p>AI has now integrated itself almost everywhere, from business and entertainment to education and healthcare. It’s almost certainly present in multiple aspects of your daily life, whether you realize it or not.</p><p>The relentless advancement of AI is hard to keep up with and easy to get lost in. As someone whose job depends on staying current with emerging technologies, I wanted to take a step back and outline the key levels at which generative AI is being used by the world’s most innovative companies.</p><h4><strong>Level 1: Classic LLM Use</strong></h4><p>The first and most common level of Generative AI is simply chatting directly with one of the many LLMs. It is highly likely that every person reading this article has conversed with a LLM in some respect, or even simply asked it a single, one-off question. The most popular of these models is Open AI’s Chat GPT, which hugely benefited from the first mover advantage, taking a mere 2 months to reach 100 million users worldwide [1]. Competition in the generative AI space is fierce, and competitors frequently dethrone each other by releasing new models that seize the number 1 space on common benchmarks. As a consumer, you should ensure that you’re maximally benefitting from this tight competition by testing out newly released models and deciding for yourself which best fits your needs instead of feeling a sense of loyalty to one company’s LLM or another.</p><p>Being able to converse with an all-knowing digital personality, capable of assuming any identity and answering any question, is certainly one of the most impactful inventions of the 21st century. However, while these tools are incredible to use for summarization, conversation, or curiosity, they do have limitations. The most prominent issue with basic LLMs is their tendency to “hallucinate”, which is when they make up a completely fabricated answer. Thankfully, simple text models only capable of chatting are mostly a thing of the past. Since late 2023, foundation model providers have equipped their LLMs with tools, added guardrails, and continued to train and improve their capabilities.</p><h4><strong>Level 2: Prompt Engineering</strong></h4><p>Sometimes it seems like LLMs confidently know everything, and our own ability is dwarfed by the speed and efficiency at which they complete tasks. In reality, when you team up and provide a helping hand to the LLM, you’re able to unlock new levels of value in ways you may not have thought possible. The first step to collaborating with an LLM is prompt engineering, which broadly refers to setting guidelines, rules, and providing examples for the LLM. There are a few different components of prompt engineering, and for maximum performance you should be utilizing all of them.</p><p>The most basic prompt you can start with is a system prompt. This is an instruction you give to your LLM at the start, which outlines its role, duty, and context. To demonstrate the effectiveness of prompt engineering, let’s imagine an example use case: You are using an LLM to summarize customer support tickets, but you want this to be done in a standardized way across the organization. The example ticket we’ll be using has the following text to be summarized:</p><blockquote>Hi team, I’ve been having ongoing issues with my account since last week. I initially reached out on Monday because my invoice for March showed an incorrect overage charge of $328. Your support rep, Jamie, said it would be resolved within 24 hours, but I never got a follow-up. I then received a service interruption email yesterday, which is really frustrating because I’ve been trying to fix this proactively. I also noticed my billing portal still says ‘past due’ even though my payment cleared on March 30th. Can someone please escalate this? I’ve already spent too much time on chat and phone support, and this is affecting our team’s usage of the product.</blockquote><p>Starting at level 0, let’s see what our assistant comes up with on its own without any sort of prompting.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*UZu-HpNQJTORzg3W" /><figcaption>No prompt.</figcaption></figure><p>You’ll notice the summary is almost as long as the ticket itself, and doesn’t seem to have any clear structure. This is not very useful to us in a real-world scenario, considering that additional support ticket summaries likely won’t look anything like this, and it’s not a very helpful summary to begin with. Turning it up a notch, let’s give the LLM a system prompt instructing it on its role and task.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*yJgkyxIhDkvR_0nC" /><figcaption>System prompt.</figcaption></figure><p>Clearly, this summarization is much more useful. It is concise and to the point, following our instruction to keep it one sentence long and including the most relevant information for an internal escalation. But what if we want it to follow a specific format or syntax to ensure consistency across our whole support team? Now we’ve arrived at the final stop in prompt engineering, called “few-shot prompting”, where we provide it with a few examples to clearly show the output format we expect.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*LHAUq6teEn2dZoCy" /><figcaption>System prompt with examples.</figcaption></figure><p>Finally, we can see that the response adheres to our optimal format and syntax. Now that our LLM knows its job and how to perform it, subsequent requests for summaries will return the exact same format without requiring new examples.</p><h4><strong>Level 3: RAG (Retrieval-Augmented Generation)</strong></h4><p>Prompt engineering allows you to customize your LLM’s outputs, formatting, and syntax, but there is still a critical piece missing from the picture. Without context, generative AI relies on the data it was trained on to retrieve information it can use to generate a response. For most simple use cases, like conversing, asking questions, or generating text this isn’t an issue because specific knowledge is not a requirement. However, in order for an LLM to answer domain-specific queries or utilize a knowledge base in its response you are required to provide it with this context. This also helps reduce hallucinations, since the LLM is relying on your hard data (documents, files, databases, etc.) to come up with its answers instead of generating them using only training data as context.</p><p>One way to provide your LLM with context is by simply including it in your queries as plain text along with your prompt. Unfortunately this is not a feasible solution at scale and becomes essentially impossible as the size of your knowledge base, documents, or other specific context expands to realistic levels which can no longer fit in the models context window. For example, imagine your collection of internal documents outlining Specific Operating Procedures (SOPs) for your organization grows to hundreds of documents. The total text between all of your SOPs would be too large to include in the context window of the LLM, and even if it weren’t you’d need to go through each document, copy the relevant text, and paste it into your prompt. It would essentially be a full time job.</p><p>Thankfully there’s already a scalable, established way for an LLM to access your specific knowledge bases. An LLM equipped with the capability to retrieve specific knowledge to use as context is called Retrieval Augmented Generation (usually just referred to as RAG). While the idea sounds simple, there are actually quite a few moving parts under the hood.</p><p>At the heart of a RAG system is a vector database (also called a vector store). This is where your documents are chunked (split into smaller pieces, like sentences or even words) and transformed into a numerical format known as embeddings. Machine learning models calculate these embeddings which enable the system to find chunks of text semantically similar to other chunks of text. For example, although the words “dog” and “puppy” are not syntactically similar, readers would understand that they are very similar in meaning and can almost be used interchangeably. A vector database captures that similarity by comparing their embeddings.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/735/0*q5dj3DaucOiPCCdD" /><figcaption><em>A visual representation of syntactic similarity in a vector database [2]</em></figcaption></figure><p>When a query is made, the RAG system searches the vector database for relevant chunks of information based on semantic similarity. These results are passed to the LLM as context, enabling it to generate responses grounded in your actual data.</p><p>Returning to our SOP example, imagine a chatbot interface that functions like any other LLM, but this time powered by a vector database filled to the brim with specific knowledge and context. Instead of answering based only on what it “knows” from training, it retrieves your organization’s official procedures and summarizes them in its response. This lets you effectively “talk to your data,” producing accurate, trustworthy outputs and dramatically reducing hallucinations. Furthermore, if we combine a RAG system with prompt engineering, we can standardize the generated output to create a production-ready chat interface.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*NYqf8TOKZtDSf9Ot" /><figcaption>A simple representation of the flow of a RAG data pipeline</figcaption></figure><h4><strong>Level 4: Agentic AI</strong></h4><p>The latest stage of generative AI, and what most modern tools are using today, is Agentic AI. Agents are semi-autonomous LLMs, equipped with tools that they understand how to use in order to achieve specific outcomes. Given a request, they will orchestrate a plan of actions to take using their tools in order to best respond to the prompt. The tools that you give your Agent can be anything you write using code, whether that’s a complex mathematical function, an API endpoint they can call, or a database query they can run. In order to best take advantage of Agents (and keep your LLMs from getting confused) it’s best to keep the scope of one individual Agent small so that it can specialize in a specific problem space. “Super Agents” equipped with a huge suite of tools sound great on paper, but they can struggle with planning out the right workflow due to the massive amount of responsibilities they have. If an agentic workflow naturally grows to be large and complex or requires many different actions, we can bring in orchestration agents. These are agents who use other agents as tools, essentially outsourcing specific pieces of work to a more low-level workflow defined by one or more subordinate agents.</p><p>To help visualize a real use case for an Agent, let’s expand on the previous chatbot example we used to demonstrate RAG. A chatbot able to recall specific knowledge from our own documents is incredibly useful for providing information and answering questions, but what if we want it to actually be able to take action on its own to solve a problem? In our hypothetical organization, we want an Agent capable of handling requests from our customers related to appointment bookings, such as cancelling or rescheduling an appointment. For starters, we’ll provide our agent with the vector database containing the specific steps we take for these processes at our organization, so it actually knows what it needs to do. Next we provide tools to run a SQL command on our database to update the booking with the new appointment date. And finally, we grant it the capability to send our customer an email confirming the updates. Et viola! A semi-autonomous agent able to perform all the steps on it’s own. In a production use case we’d want to test and evaluate it rigorously before deploying it in customer facing scenarios. It’s also worth noting that this is a very high-level view of our agentic workflow and will vary depending on the tools and software you’re using to create your Agent.</p><p>As the space continues to evolve, we’ll see new layers of generative AI, rising and falling foundation models, and startups redefining what’s possible. The organizations and individuals that succeed in harnessing this new wave will unlock value that just a few years ago seemed so far out of reach.</p><p><strong>Sources</strong></p><ol><li><a href="https://businessday.ng/technology/article/chatgpt-is-fastest-app-to-hit-100m-users-in-history/">https://www.google.com/url?sa=i&amp;url=https%3A%2F%2Fbusinessday.ng%2Ftechnology%2Farticle%2Fchatgpt-is-fastest-app-to-hit-100m-users-in-history%2F&amp;psig=AOvVaw1LUxamuoq56HdV5RPo0FIa&amp;ust=1750460448663000&amp;source=images&amp;cd=vfe&amp;opi=89978449&amp;ved=0CBQQjRxqFwoTCIiisbvL_o0DFQAAAAAdAAAAABBR</a></li><li><a href="https://weaviate.io/assets/images/image10-ebe747ac9f2e03dba758f1ed3ea7e82c.jpg">https://weaviate.io/assets/images/image10-ebe747ac9f2e03dba758f1ed3ea7e82c.jpg</a></li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*I0478Hountx4GWvk.png" /></figure><p>This story is published on <a href="https://generativeai.pub/">Generative AI</a>. Connect with us on <a href="https://www.linkedin.com/company/generative-ai-publication">LinkedIn</a> and follow <a href="https://www.zeniteq.com/">Zeniteq</a> to stay in the loop with the latest AI stories.</p><p>Subscribe to our <a href="https://www.generativeaipub.com/">newsletter</a> and <a href="https://www.youtube.com/@generativeaipub">YouTube</a> channel to stay updated with the latest news and updates on generative AI. Let’s shape the future of AI together!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*oU9PjvkROGIadPQh.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b6ddbfa407e8" width="1" height="1" alt=""><hr><p><a href="https://generativeai.pub/from-prompts-to-agents-b6ddbfa407e8">From Prompts to Agents</a> was originally published in <a href="https://generativeai.pub/">Generative AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Databricks FE Launchpad Program 3-Month Update]]></title>
            <link>https://medium.com/@finpiroth/databricks-fe-launchpad-program-3-month-update-2b8ecbf2e486?source=rss-5d89cb499422------2</link>
            <guid isPermaLink="false">https://medium.com/p/2b8ecbf2e486</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[career-development]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[databricks]]></category>
            <dc:creator><![CDATA[Finlay Piroth]]></dc:creator>
            <pubDate>Fri, 09 May 2025 19:25:08 GMT</pubDate>
            <atom:updated>2025-05-11T20:35:48.146Z</atom:updated>
            <content:encoded><![CDATA[<h3>Databricks FE Launchpad Program</h3><h4><em>An inside perspective from Databricks inaugural cohort of solutions engineers</em></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LkDkqDbh7cnWIMX1Js5H2Q.jpeg" /><figcaption>One of the many pieces of Databricks swag I’ve collected</figcaption></figure><p>The recent AI boom has captivated the interest of businesses, with 83% reporting AI integration as a top priority [1]. But while executives eagerly explore ways to embed AI across their operations, many employees are left uneasy, concerned that their roles could soon be replaced. With companies racing to innovate and employees anxious about their future, a unique opportunity has emerged which addresses the needs of both.</p><p>Organizations that invest in the next generation of talent and offer training in cloud, data, and AI are not only building a future-ready workforce, but also gaining a competitive edge through a tailor-made, highly skilled talent pool. Databricks is leading the charge into enterprise data and AI, scooping up top talent and up-skilling early career employees to be the next generation of industry experts. I’m proud to be part of the inaugural Field Engineering (FE) Launchpad program, a hands-on initiative designed to train early-career professionals to be solutions engineers specializing in data and AI. With three months now behind us, here’s what we’ve achieved and what lies ahead.</p><p>The “cloud” serves as a backbone for most modern businesses, and often refers to a collection of resources offered through a pay-as-you-go model by AWS, Azure, or a variety of other providers (we call these “hyperscalers”). It is an incredibly vast field, demanding in-depth knowledge of networking, computational infrastructure, and database concepts to name a few. Mastery of these technologies is a highly sought after credential, and most hyperscalers offer a few different certifications if you are able to pass their exams. One of our first goals in FE Launchpad was to obtain two cloud certifications — we chose between AWS Certified Solutions Architect Associate (SAA-C03) or Azure Administrator Associate (AZ-104), and were also required to secure the Databricks Certified Data Engineer Associate certification. Thankfully, Databricks offers its employees a generous budget for professional development, which allowed us to subscribe to advanced courses and unlock practice exams to prepare us for the real deal. Not only was the knowledge we gained through this experience incredibly useful for our day-to-day work, but we have also all achieved a valuable stamp of approval proving our skill and expertise in the industry.</p><p>Equally important to obtaining cloud certifications is the actual learning, training, and application of domain knowledge through hands-on projects and experiences. For the past 3 months, each week we tackled a new topic (for example “Data Warehousing”) and corresponding trainings in the form of videos, labs, or readings, and a list of hands-on exercises to complete at our own pace throughout the week. Often accompanying these assignments were guest speakers from the company with expertise in the domain, followed by leadership chats with executives at Databricks. This provided our cohort the opportunity to ask questions about the company, specific topics, or anything else we were interested in hearing from these leaders and experts. Lastly, as part of our practice in speaking and presenting, we presented in pairs to the rest of our group each Friday on a specific topic within that week’s domain.</p><p>Our cohort consists of 20 early career professionals with varying backgrounds from software and data engineers to tech consultants or salespeople, but the one thing we all have in common is that we were seeking a new path where we could utilize both our interpersonal skills and our technical knowledge. All 20 of us started on the same day across 3 different office locations in the US, and although we primarily interact in-person with our teammates who are in the same office, we collaborate remotely with those in other offices as well. Together we prepared presentations for weekly tech talks to the rest of our cohort, we collaborated on exercises and assigned training, and we studied for our cloud certifications. One of the most memorable moments in FE Launchpad was during our in-person bootcamp, during which we were all flown out to one of the company’s headquarters and stayed in a hotel nearby. Our workdays were packed with intense training, guest speakers, and team building exercises, while after work we had team dinners, happy hours, and other activities for us to network and socialize. The week-long bootcamp culminated in a capstone project which we worked on in groups of 3, presenting our projects on the final day of bootcamp and receiving feedback from the manager and peers who we presented to.</p><p>As our initial 3 months of training, exercises, and certification studying came to a close, we began the transition into our actual customer-facing duties. Based on our background and interests, we have each been assigned into different Business Units (BU), each of which primarily encompasses the same classification of customer, for example, “Startups” or “Health and Life Science”. Within our BU we were assigned a mentor, whom we would go on to learn from and shadow in their customer calls to get a feel for the type of work we eventually would be doing. As someone who recently transitioned from software engineering, stepping into a client-facing role is both exciting and a bit daunting. But the depth of preparation provided by FE Launchpad gives me confidence that I’m ready to contribute, and I’m eager to start delivering impact.</p><p><strong>Sources</strong></p><ol><li><a href="https://www.nu.edu/blog/ai-statistics-trends/#:~:text=83%25%20of%20companies%20reported%20that,strategies%20is%20a%20top%20priority.&amp;text=52%25%20of%20employed%20respondents%20are%20worried%20AI%20will%20replace%20their%20jobs.&amp;text=The%20manufacturing%20sector%20will%20likely,of%20%243.8%20trillion%20by%202035">https://www.nu.edu/blog/ai-statistics-trends/#:~:text=83%25%20of%20companies%20reported%20that,strategies%20is%20a%20top%20priority.&amp;text=52%25%20of%20employed%20respondents%20are%20worried%20AI%20will%20replace%20their%20jobs.&amp;text=The%20manufacturing%20sector%20will%20likely,of%20%243.8%20trillion%20by%202035</a>.</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2b8ecbf2e486" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Best Tech Job You’ve Never Considered]]></title>
            <link>https://medium.com/@finpiroth/the-best-tech-job-youve-never-considered-d0ef72b8a219?source=rss-5d89cb499422------2</link>
            <guid isPermaLink="false">https://medium.com/p/d0ef72b8a219</guid>
            <category><![CDATA[career-change]]></category>
            <category><![CDATA[solutions-architecture]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[software-engineering]]></category>
            <dc:creator><![CDATA[Finlay Piroth]]></dc:creator>
            <pubDate>Fri, 28 Feb 2025 21:57:05 GMT</pubDate>
            <atom:updated>2025-02-28T21:57:05.146Z</atom:updated>
            <content:encoded><![CDATA[<h4>How I Transitioned from Software to Solutions Engineering</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pGhGpdBYChwZii4GUxGCjg.png" /></figure><p>One month ago, I did the unthinkable: I walked away from one of the highest-paying tech companies in Silicon Valley, slipping out of my golden handcuffs in search of something — anything — more fulfilling. It felt like a scene straight out of <em>The Matrix</em>, staring down the choice between the red pill and the blue pill. The blue pill meant comfort, stability, and a pending stock grant worth as much as my life savings. The red pill meant uncertainty, risk, and waving goodbye to a life-changing bonus — all in pursuit of a career that didn’t make me question my existence daily.</p><p>For years, I had chased the dream: a coveted software engineering role at a top tech company straight out of college. Sleepless nights, caffeine-fueled coding marathons, cramming for finals — every grueling moment had finally paid off. I was surrounded by brilliant engineers, tackling some of the world’s most complex systems, and enjoying compensation that would make my college self weep with joy. But, as with any dream, the reality was far more complicated than the glossy LinkedIn posts made it seem.</p><p>The day-to-day of a software engineer involved things my university had conveniently left out of the syllabus. The joy of writing elegant code was quickly overshadowed by week-long, round-the-clock on-call rotations where my peaceful slumber could be shattered by a 2 AM page about a production outage. Performance reviews felt like judgment day, where my future rested entirely on my ability to churn out code, regardless of how burnt out I felt. And human interaction? Limited to a 10-minute stand-up meeting and a brief lunch break where I attempted to recall my left-behind conversational skills. Most of my waking hours were spent in isolation, AirPods in, staring at my screen, wondering if this was all there was.</p><p>As the pressure mounted, I started to suspect I had made a terrible mistake. But quitting wasn’t an option — at least, not without a plan. The tech industry was experiencing a once-in-a-generation downturn, with mass layoffs and a job market so bleak it made dystopian fiction look optimistic. So, I gritted my teeth and kept grinding, pouring months into job applications, interviews, and soul-crushing rejections. Until, against all odds, I had a breakthrough: an offer to join Databricks as a Solutions Engineer.</p><p>The interview process was no joke — two months of grueling technical and behavioral screenings, culminating in a high-stakes 30-minute presentation to a panel of hiring managers. But when the offer came through, I knew I had found my escape hatch.</p><p>Solutions engineering is the best of both worlds: the intellectual challenge of technical problem-solving, combined with the energy of real human interaction. Instead of toiling away in solitude, I collaborate directly with customers, helping them design and implement data solutions using Databricks. I’d still be flexing my technical muscles, but in a way that felt dynamic, engaging, and — dare I say it — fun. It was a chance to put my software engineering background to use without being chained to a desk, measuring my worth in lines of code.</p><p>Now, one month into my new role, I can confidently say: this was the right move. I’m part of Databricks’ newly minted Inside SA program (also referred to as the Field Engineering Launchpad), a cohort of 20 ambitious career-switchers, majority of whom are making the leap from software or data engineering to solutions engineering. We’re deep in training, learning from seasoned mentors, and working toward certifications that will shape us into the next generation of solutions architects. It’s exhilarating, rewarding, and — perhaps most importantly — I no longer wake up in a cold sweat dreading another day of debugging mysterious production failures at ungodly hours.</p><p>In short, I made a career move that finally feels right— one that challenges me, engages me, and makes work exciting again. And I wouldn’t trade that for anything.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d0ef72b8a219" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>