<?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 Hailey Jun on Medium]]></title>
        <description><![CDATA[Stories by Hailey Jun on Medium]]></description>
        <link>https://medium.com/@hailey_dataforsocialgood?source=rss-b75ccbbcad05------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*x8s1BsGi3aoIgYgiKIE44A.jpeg</url>
            <title>Stories by Hailey Jun on Medium</title>
            <link>https://medium.com/@hailey_dataforsocialgood?source=rss-b75ccbbcad05------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 21:58:15 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@hailey_dataforsocialgood/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[Is Deepfake good or bad?]]></title>
            <link>https://medium.com/@hailey_dataforsocialgood/is-deepfake-good-or-bad-15b83297cdc2?source=rss-b75ccbbcad05------2</link>
            <guid isPermaLink="false">https://medium.com/p/15b83297cdc2</guid>
            <dc:creator><![CDATA[Hailey Jun]]></dc:creator>
            <pubDate>Thu, 14 Aug 2025 02:54:24 GMT</pubDate>
            <atom:updated>2025-08-15T03:42:54.043Z</atom:updated>
            <content:encoded><![CDATA[<h3>Deepfakes: A Double-Edged Sword in the Age of AI</h3><p>A recent session from “AI for Good,” featuring insights from Michaela A. Ternasky-Holland and Sam Gregory, delved into the complex world of deepfake technology. Often cited as a prime example of tech’s potential to harm society, deepfakes underwent a thorough examination from both beneficial and detrimental perspectives.</p><h3>Can Deepfakes Be a Force for Good?</h3><p>The discussion first explored the positive applications of deepfakes. A notable project highlighted was <strong>Kapwa</strong> by Michaela Ternasky-Holland. This initiative used deepfake technology to celebrate Philippine independence, creating an immersive “what if” scenario for users who aren’t Filipino, allowing them to experience a connection to the culture.</p><blockquote><em>Kapwa is not only a visceral representation of our struggle to understand what is truly Filipino culture but also a… michaelaternaskyholland.com</em></blockquote><p><a href="https://michaelaternaskyholland.com/case-studies/kapwa">Kapwa - Michaela Ternasky-Holland</a></p><p>This compelling example demonstrates that AI, specifically deepfake technology, can play a critical role in the art industry. But this raises a fundamental question: <strong>What is the standard for art in the age of AI?</strong></p><h3>Defining Art in a New Era</h3><p>The dictionary defines art as “the expression or application of human creative skill and imagination, typically in a visual form such as painting or sculpture, producing works to be appreciated primarily for their beauty or emotional power.”</p><p>Consider the profound shock and industry-wide disruption caused by the transition from traditional drawings to photography. Photography was arguably seen by some as a mere machine automation of human creative endeavor, seemingly devoid of the true creativity and meaning found in painting.</p><p>So, is AI art, truly art? Let’s revisit the definition. First, it is an <strong>expression</strong> — if the person behind the prompt has a meaning to convey through the piece. Second, it is a <strong>visual form</strong>. But third, can it truly be considered a part of <strong>human creativeness</strong>? The artist, in this context, is the prompter.</p><p>The key difference between traditional photography and AI art lies in the machine’s role. In photography, the human controls the frame; the camera performs a relatively passive task of precisely capturing what the photographer sees and intends to print. Thus, the act of framing the world through a lens and dictating its reproduction is clearly seen as an intentional and artistic action.</p><p>By contrast, AI plays a more active role in AI art. There’s an inherent uncertainty in what the AI will produce based on a given prompt. While the human provides guidance, the machine’s share in the creative output of AI art is demonstrably larger than what we’ve seen in photography.</p><p>However, regardless of the degree of machine intervention in what is essentially a human-computer collaborative art form, the prompt itself is undeniably a <strong>creative process</strong>. Therefore, AI art can indeed be regarded as a legitimate form of art.</p><h3>The Dark Side: Malicious Deepfakes</h3><p>Moving to the other side of the coin, it’s undeniable that deepfake technology can create significant chaos for societal information management.</p><p><a href="https://www.reuters.com/fact-check/real-video-pikachu-protester-turkey-followed-by-ai-fakes-online-2025-04-03/">Fact Check: Real video of Pikachu protester in Turkey followed by AI fakes online</a></p><blockquote><em>Fact Check: Real video of Pikachu protester in Turkey followed by AI fakes online Several AI-generated photos of Pokemon character Pikachu were widely shared online after a verified video of a person… [suspicious link removed]</em></blockquote><p>Deepfakes can easily be weaponized to spread political misinformation and disrupt security systems. Furthermore, developing a robust watermarking system for online photos to ensure their authenticity remains a significant challenge.</p><p>The problem of misinformation becomes even more intricate when deepfake images are partly based on reality. For instance, in the protest example linked above, a Pikachu mascot was actually present. However, individuals then used AI to generate further provocative images of Pikachu at the protest, mimicking the true photo, which made it incredibly confusing for readers to discern fact from fiction.</p><h3>What Do We Need Moving Forward?</h3><h3>1. Clear Boundaries and Consensus</h3><p>First, we urgently need to establish clear boundaries for what is acceptable and what is not. While deepfakes can serve artistic expression, they should never be used to confuse people or spread false political information. Currently, there’s no clear social consensus on what constitutes tolerable behavior. This requires not just social norms, but also comprehensive legal frameworks and regulations covering copyright and AI behavior. We’re already seeing steps in this direction, with platforms like Instagram and YouTube requiring creators to declare AI-generated images that resemble reality.</p><h3>2. Robust Detection Systems</h3><p>Second, we need highly effective detection systems. Despite any legal or regulatory frameworks, there will always be individuals or groups who circumvent the rules. For this, we must invest in building robust AI-generated content detectors that can keep pace with the evolving sophistication of deepfake technology.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=15b83297cdc2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[[Paper Review] Unmasking Financial Crime: How Deep Learning Aids Fraud Detection in Brazilian…]]></title>
            <link>https://medium.com/@hailey_dataforsocialgood/paper-review-unmasking-financial-crime-how-deep-learning-aids-fraud-detection-in-brazilian-624bd8d5cf78?source=rss-b75ccbbcad05------2</link>
            <guid isPermaLink="false">https://medium.com/p/624bd8d5cf78</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[machinelerning]]></category>
            <dc:creator><![CDATA[Hailey Jun]]></dc:creator>
            <pubDate>Thu, 29 May 2025 05:54:47 GMT</pubDate>
            <atom:updated>2025-05-29T05:54:47.229Z</atom:updated>
            <content:encoded><![CDATA[<h3>[Paper Review] Unmasking Financial Crime: How Deep Learning Aids Fraud Detection in Brazilian Exports</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/768/1*XX0Lcs7n1BYhRXxedMq3cQ.jpeg" /><figcaption>Photo: Brazilian Federal Police</figcaption></figure><p>This is a review of the paper — <a href="https://ieeexplore.ieee.org/document/7838276">https://ieeexplore.ieee.org/document/7838276</a></p><p>Money laundering and financial fraud pose significant threats to global economies. The complex nature of international trade, like Brazil’s vast export market reaching over 200 countries, can unfortunately become a conduit for illicit financial activities. Sifting through a colossal amount of transaction data to distinguish legitimate operations from suspicious ones presents a monumental challenge. This is where cutting-edge technology steps in. A novel approach utilizing deep learning, specifically autoencoders, offers a promising solution to support fraud investigation in Brazilian exports and bolster anti-money laundering (AML) efforts. The goal is to pinpoint exporting companies exhibiting anomalous behavior that warrants further scrutiny.</p><h3>The Brazilian Push Against Money Laundering</h3><p>Brazil has taken significant strides in combating money laundering. The introduction of Law 12.683/2012 marked a major advancement in the nation’s AML framework. This law strengthened penalties for money laundering crimes and, crucially, expanded the definition of predicate offenses. This means that proceeds from any crime or misdemeanor, such as illegal gambling or exploitation of slot machines, could be targeted as money laundering. Organized crime often uses money laundering to give a veneer of legitimacy to their illicit earnings. The 2012 law enhanced the state’s ability to fight such organized crime. It also broadened the range of entities required to report suspicious transactions to the Council for Financial Activities Control (COAF) and significantly increased fines for non-compliance.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ytTIFMMYh2KTv1l-VIHpiQ.png" /></figure><p>It’s important to note that anomalies detected by systems like the one discussed are not immediate proof of guilt. Instead, they serve as critical flags, guiding authorities to entities that require further investigation through audits, legal actions, or cross-verification of data.</p><h3>A Structured Approach to Uncovering Anomalies: CRISP-DM</h3><p>The methodology behind this fraud detection system follows the Cross Industry Standard Process for Data Mining (CRISP-DM). This structured approach ensures a comprehensive and effective data analysis process, encompassing:</p><ul><li><strong>Business Understanding</strong></li><li><strong>Data Understanding</strong></li><li><strong>Data Preparation</strong></li><li><strong>Modeling</strong></li><li><strong>Evaluation</strong></li><li><strong>Deployment</strong></li></ul><p>The system leverages a rich dataset sourced from Brazil’s foreign trade secretariat of federal revenue. This data encompasses various categories:</p><ul><li><strong>Registration Information:</strong> Exporter identification and trade operation licenses.</li><li><strong>Trade Information:</strong> Details like export volume, amounts, product classifications (NCM), and origins/destinations.</li><li><strong>Tax Collection:</strong> Information on taxes and fees levied during export activities.</li><li><strong>Financial Transactions:</strong> Data from Brazilian financial institutions, including transaction amounts and bank accounts.</li><li><strong>Withholding Tax:</strong> Taxes withheld by companies for services.</li><li><strong>Employee Data:</strong> Social security contributions made by companies for their employees.</li><li><strong>Electronic Invoices:</strong> Data from electronic invoices for goods purchased and sold for export.</li><li><strong>Accessory Obligations:</strong> Compliance with various declarations submitted to tax authorities.</li><li><strong>Inspection History:</strong> Past tax and customs inspection records for export companies.</li><li><strong>Other Surveillance Data:</strong> Information from other monitoring activities related to export companies.</li></ul><p>In total, the model was developed using approximately 819,990 transaction records.</p><h3>The Heart of Detection: Autoencoders Explained</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*l3bVHLhyoXOu_gkOFqNXQw.png" /><figcaption>Auto Encoder Structure — <a href="https://wikidocs.net/202070">https://wikidocs.net/202070</a></figcaption></figure><p>At the core of this anomaly detection system lies an <strong>autoencoder</strong>, a type of artificial neural network designed to learn efficient representations of data. An autoencoder consists of two main parts:</p><ol><li><strong>Encoder:</strong> This part compresses the input data into a lower-dimensional representation, often called the “latent space” or “bottleneck.” The key is that this compressed form, h=f(x), captures the most important features of the data.</li><li><strong>Decoder:</strong> This part attempts to reconstruct the original input data from the compressed representation: r=g(h).</li></ol><p>The entire process can be visualized as x⇒f(x)=h⇒g(h). The autoencoder is trained to minimize the <strong>reconstruction loss</strong> (or error), which is the difference between the original input (x) and the reconstructed output (g(h)). Think of it like creating a concise summary of a book (encoding) and then trying to rewrite the original book using only that summary (decoding). The better the summary and reconstruction process, the lower the error.</p><p>This training is <strong>self-supervised</strong>, meaning it learns the inherent structure and characteristics of the data without needing explicit labels for “normal” or “abnormal” instances.</p><h3>How Autoencoders Spot the Odd Ones Out</h3><p>The power of autoencoders in anomaly detection comes from training them predominantly on “normal” data. The network learns the typical patterns and relationships between features in these normal instances. Consequently:</p><ul><li>When <strong>normal data</strong> is fed into the trained autoencoder, it can be compressed and reconstructed with a <strong>low reconstruction error</strong>.</li><li>When <strong>anomalous data</strong> (which deviates significantly from the learned normal patterns) is input, the autoencoder struggles to reconstruct it accurately, resulting in a <strong>high reconstruction error</strong>.</li></ul><p>The Mean Square Error (MSE) is often used as the reconstruction error metric:</p><ul><li>A higher MSE value signifies that a particular record is more anomalous compared to the patterns the autoencoder learned from the data.</li></ul><p>Interestingly, the study found that even the representation of data in the autoencoder’s middle layer (the latent space) can reveal anomalies. Highly anomalous data points often appear as outliers even in this compressed dimensional space, distinct from the cluster of normal data points. A practical aspect noted was that running too many training epochs could lead to overfitting; around 50 epochs with the ReLU activation function was found to be optimal.</p><h3>Evaluating Anomalies: Pinpointing Companies for Review</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/748/1*PeXv70q8y6bxQ_vJkmPEZw.png" /></figure><p>The evaluation of the model focuses on identifying exporting companies that exhibit the most anomalous transaction patterns. By calculating the MSE for each company’s aggregated transaction data, the system can rank them by their anomaly score.</p><p>The distribution of these MSE values, when sorted, often shows a distinct change in behavior for the highest-scoring records. For instance, in this study, a clear deviation was observed for the top 20 most anomalous records out of 170 highlighted cases, indicating a strong signal for potential irregularities. These companies would then become the prime candidates for more detailed investigation by authorities.</p><h3>Paving the Way for Smarter Financial Oversight</h3><p>The application of deep learning, particularly autoencoders, provides a powerful tool in the ongoing fight against financial crime. By learning the intricate patterns of normal behavior within massive datasets of Brazilian export transactions, this approach offers an efficient way to flag suspicious activities that might otherwise go unnoticed. It represents a significant step forward in leveraging artificial intelligence to support fraud investigation and strengthen anti-money laundering initiatives, ultimately contributing to a more secure and transparent trade environment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/780/1*K0JT5kP1swpEi__JkZSy6w.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=624bd8d5cf78" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Tackling Waterborne Diseases: Status and How to Overcome with Technology/Business is]]></title>
            <link>https://medium.com/@hailey_dataforsocialgood/tackling-waterborne-diseases-status-and-how-to-overcome-with-technology-business-is-8297d440b6fe?source=rss-b75ccbbcad05------2</link>
            <guid isPermaLink="false">https://medium.com/p/8297d440b6fe</guid>
            <category><![CDATA[tech4good]]></category>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[sdgs]]></category>
            <dc:creator><![CDATA[Hailey Jun]]></dc:creator>
            <pubDate>Fri, 23 May 2025 09:21:15 GMT</pubDate>
            <atom:updated>2025-05-23T09:21:15.204Z</atom:updated>
            <content:encoded><![CDATA[<h3>Tackling Waterborne Diseases: Status and How to Overcome with Technology/Business</h3><p>Access to clean water and proper sanitation is a fundamental human right, yet millions worldwide still lack these basic necessities. This deficiency leads to a pervasive and deadly problem: waterborne diseases. Let’s dive into the critical issues surrounding water quality, the diseases it causes, and the innovative solutions being deployed to combat this global challenge.</p><h3>The State of Our Water</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/908/1*2aBAMGK0OTl7OTG6EGYVTA.png" /><figcaption>Figure 1. WWF Water Risk Filter water quality risk map<br>based on World Bank’s water quality dataset — <a href="https://wbwaterdata.org/">https://wbwaterdata.org/</a></figcaption></figure><p>A glance at the global water quality map reveals a stark reality: many regions, particularly in Africa and parts of Asia, face “very high” water quality risk. This isn’t just an environmental concern; it’s a public health crisis. The lack of safe water directly contributes to the spread of numerous illnesses.</p><h3>The Silent Killers: Common Waterborne Diseases</h3><p>Waterborne diseases are caused by pathogenic microorganisms transmitted through contaminated water. The most common culprits include:</p><ul><li><strong>Diarrhea:</strong> Often the most prevalent and deadly.</li><li><strong>Cholera:</strong> A severe bacterial infection causing acute watery diarrhea.</li><li><strong>Typhoid Fever:</strong> A bacterial infection that can lead to high fever, fatigue, and abdominal pain.</li><li><strong>Hepatitis A:</strong> A viral liver infection.</li><li><strong>Giardiasis:</strong> An intestinal infection caused by a parasite.</li><li><strong>Malaria:</strong> While primarily mosquito-borne, its prevalence can be linked to stagnant water bodies often associated with poor drainage and sanitation.</li></ul><p>These diseases primarily transmit through contaminated water, poor sanitation, lack of hygiene, and the fecal-oral route.</p><h3>The Devastating Impact: More Than Just Sickness</h3><p>The consequences of unsafe water, sanitation, and hygiene (WASH) services are staggering:</p><ul><li><strong>Preventable Deaths:</strong> In 2019 alone, 1.4 million deaths could have been prevented with safe WASH.</li><li><strong>Morbidity:</strong> 74 million Disability-Adjusted Life Years (DALYs) could have been prevented. DALYs measure the total number of healthy years lost due to illness, disability, or early death.</li><li><strong>Diarrhea’s Toll:</strong> Over 1 million people died from diarrhea due to unsafe WASH in 2019, accounting for 69% of all diarrhea deaths. Diarrheal disease is tragically the third leading cause of death in children under 5, claiming around 443,832 young lives annually.</li></ul><h3>Open Defecation: A Major Contributor</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/676/1*t1IZIQfR1Rzy7_CaP-kkjg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*weCC8lQdA8e7n0--6OtOMw.png" /><figcaption>Mortality rate attributed to exposure to unsafe WASH services — Source : WHO</figcaption></figure><p>One of the most significant challenges in sanitation is open defecation, where people relieve themselves in fields, bushes, or open spaces. As of 2022, countries like Eritrea (67%), Niger (65%), and Chad (63%) show alarmingly high rates. Notably, 17 of the top 20 countries practicing open defecation are African.</p><p>Factors contributing to this practice include:</p><ul><li>Lack of public awareness on sanitation/primitive lifestyles.</li><li>Lack of adequate water supply.</li><li>Absence of necessary technology for appropriate toilet models, especially in rural areas.</li><li>Poor maintenance of existing facilities.</li><li>Heavy financial requirements to tackle the issue.</li></ul><h3>Why the Struggle Persists: Underlying Issues</h3><p>The challenge of providing clean water and sanitation is multifaceted, stemming from various interconnected problems:</p><ul><li><strong>Geopolitical/Environmental Factors:</strong> Countries like Niger, being landlocked and 80% desert with low and irregular annual rainfall and high evaporation rates, face extreme water scarcity.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/562/1*zQ0LubRhVY-6bYf1WFTIEQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/890/1*EILiYU9ujymEENvuuNlIdQ.png" /><figcaption>Death by diarreah — 2004</figcaption></figure><ul><li><strong>Economic Development Issues:</strong> High poverty rates (e.g., 52% in Niger), economies heavily reliant on agriculture, and significant impacts from climate change hinder progress. Political instability also makes infrastructure investment difficult. There’s a severe lack of infrastructure investment, with an estimated annual investment gap of $225 million (or $460 million total deficit) in Niger. Over 80% of the population lives in rural areas, requiring more investment for infrastructure.</li><li><strong>Population Structure &amp; Social Problems:</strong> Rapid population growth (Niger’s population doubled between 1990–2008, with a current 4% annual growth rate) strains existing water and sanitation infrastructure. Cultural practices, such as 71% of the population practicing open defecation, and social discrimination against women (who bear the burden of water collection), exacerbate the problem, leading to time poverty and lack of education.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/694/1*nRnj1girKbKp_lPygdP5BQ.png" /><figcaption>Source — Concern net</figcaption></figure><ul><li><strong>Political &amp; Historical Issues:</strong> Water resource management often requires complex international cooperation, as seen with the Niger River Commission and Niger Basin Authority, which have historically been inefficient.</li></ul><h3>Hope on the Horizon: Innovative Solutions</h3><p>Despite the immense challenges, various technological applications and initiatives are offering solutions:</p><p><strong>Case 1: Drink Well (ATM-based Water Dispensing)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/872/1*91mEE0GotAbYHPfQmC-E8g.png" /><figcaption>Source — Drink well</figcaption></figure><ul><li>This initiative provides clean drinking water through ATM-like dispensers.</li><li>It’s a solution developed by a company founded by Johns Hopkins University public health graduates.</li></ul><p><strong>Case 2: mWater (Data-Driven Water Management)</strong></p><ul><li>mWater is a free, scalable technology platform that enables data-driven management by governments, utilities, water authorities, NGOs, and the private sector.</li><li>It includes a surveyor mobile app, a data portal for managing and visualizing data, and “Solstice” for upgrading human resource capacity.</li><li>It supports data collection and analysis, surveys, and water point mapping.</li><li><strong>Real-world Impact:</strong> During Cyclone Idai in Malawi, mWater helped inform water point rehabilitation by identifying flooded and contaminated wells, providing crucial data for response efforts. In Uganda, it aids site selection for water infrastructure projects by identifying vulnerable areas based on water resource data.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/852/1*l6Ac_1XJRljhy7bbtt6fVQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/798/1*2vEy_TegTP3aes2uFti4ZA.png" /><figcaption>Source — M-water website</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RfSsLABabWyNPtbugslHIw.png" /><figcaption>Using M-water for site selection of water infrastructure</figcaption></figure><p><strong>Case 3: Zenysis (Public Health Monitoring &amp; Data Analysis)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rk6AyUjlf_ZtvVR_BOZfeA.png" /><figcaption>Source — Zenysis website</figcaption></figure><ul><li>Zenysis provides a platform for public health monitoring, data analysis, and forecasting.</li><li>It involves partnerships with various countries and organizations.</li><li><strong>Examples:</strong> Their work includes supporting Pakistan’s Zero Dose Challenge (expanding vaccination access) and South Africa’s HIV Response (expanding access to HIV treatment).</li></ul><h3>Moving Forward: A Collective Responsibility</h3><p>The journey towards universal access to clean water and sanitation is ongoing. It requires a concerted effort from governments, international organizations, local communities, and innovators. By understanding the complexities of the problem and supporting effective solutions, we can collectively work towards a future where safe WASH services are a reality for everyone, preventing countless illnesses and saving lives.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8297d440b6fe" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Agentic Retrieval-Augmented Generation (RAG)]]></title>
            <link>https://medium.com/@hailey_dataforsocialgood/agentic-retrieval-augmented-generation-rag-b9317f09eb06?source=rss-b75ccbbcad05------2</link>
            <guid isPermaLink="false">https://medium.com/p/b9317f09eb06</guid>
            <dc:creator><![CDATA[Hailey Jun]]></dc:creator>
            <pubDate>Tue, 22 Apr 2025 08:32:32 GMT</pubDate>
            <atom:updated>2025-04-22T08:33:17.356Z</atom:updated>
            <content:encoded><![CDATA[<p>This post is based on this <a href="https://arxiv.org/abs/2501.09136">paper</a></p><p>As Large Language Models (LLMs) continue to evolve, they face limitations with outdated information and context management. Retrieval-Augmented Generation (RAG) helps overcome these challenges by incorporating external knowledge sources. This article explores the next evolution: Agentic RAG, which combines agent-like capabilities with traditional RAG systems for more powerful, autonomous AI solutions.</p><h3>1. Introduction</h3><p>Despite the remarkable capabilities of Large Language Models (LLMs), they face significant limitations: their knowledge is static as of their training cutoff, and they struggle with complex reasoning tasks. Retrieval-Augmented Generation (RAG) emerged as a technique to address these challenges by connecting LLMs to external knowledge sources.</p><p>However, traditional RAG systems still face obstacles when handling complex queries requiring multi-step reasoning, tool use, or domain-specific knowledge. This is where <strong>Agentic RAG</strong> comes into play, representing the next evolution in LLM-powered systems.</p><h3>What is Agentic RAG?</h3><p>Agentic RAG introduces agent-like capabilities to RAG systems, enabling them to:</p><ul><li><strong>Reflect:</strong> Evaluate their own outputs and reasoning processes</li><li><strong>Plan:</strong> Break down complex tasks into manageable steps</li><li><strong>Use Tools:</strong> Interact with APIs and external services to gather information</li><li><strong>Collaborate:</strong> Work with other agents for complex problem-solving</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*C92MUdKpOUr3cVwo.png" /><figcaption>Example : source —miro medium</figcaption></figure><p>Figure 1: Agentic RAG combines traditional RAG with agent-like capabilities for enhanced performance</p><h3>2. Retrieval-Augmented Generation (RAG)</h3><h3>2.1 The Concept of RAG</h3><p>Retrieval-Augmented Generation enhances LLMs by enabling access to external knowledge sources. Unlike traditional LLMs with static knowledge from their training data, RAG dynamically retrieves relevant information from databases, documents, or APIs as needed.</p><blockquote>“RAG can be defined as a technique that enhances LLM capabilities by retrieving information from external knowledge sources (documents, databases, APIs) before generating a response.”</blockquote><h3>2.2 Key Components of RAG</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*JFZhhhkpnplCA0KZ.png" /><figcaption>Example : source — miro.medium</figcaption></figure><p>ComponentDescription<strong>Retrieval</strong>Finding relevant information from knowledge sources (databases, documents, web content)<strong>Augmentation</strong>Integrating the retrieved information, preprocessing it for the LLM<strong>Generation</strong>LLM using the augmented context to produce a response</p><p>Figure 2: The three main components of a RAG system and their interactions</p><h3>2.3 Evolution of RAG Systems</h3><p>RAG TypeCharacteristicsLimitations</p><ol><li><strong>Naive RAG<br></strong>Basic keyword matching (TF-IDF, BM25) <br>Limited semantic understanding, misses context</li></ol><p><strong>2. Advanced RAG<br></strong>Dense embeddings, neural rerankersStill lacks reasoning capabilities</p><p><strong>3. Modular RAG<br></strong>Specialized components for different tasksComplexity in component orchestration</p><p><strong>4. Graph RAG<br></strong>Knowledge represented as interconnected graphsComplex to implement and maintain</p><p><strong>5. Agentic RAG<br></strong>Agent-like capabilities for autonomous operationOur current frontier!</p><h3>2.3.1 Naive RAG</h3><p>Early RAG systems used simple keyword matching techniques like TF-IDF or BM25 to retrieve relevant documents from a dataset. These approaches primarily relied on lexical matching, struggling with semantic understanding.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Gf8ua5QaiDAt5kMGkrvSLw.png" /><figcaption>Architecture of a Naive RAG system using keyword matching</figcaption></figure><p><strong>Limitations</strong> include poor handling of synonyms and inability to understand the context behind queries.</p><p><strong>Example:</strong> For the query “What are the effects of vitamin D deficiency?”, a Naive RAG system might miss documents discussing “sunshine vitamin shortage” if these exact keywords aren’t present.</p><h3>2.3.2 Advanced RAG</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XJSjADvFWKGYXoxRMorV4A.png" /><figcaption>Advanced RAG</figcaption></figure><p>Advanced RAG systems use dense embedding models to capture semantic meaning, creating vector representations of documents and queries to find relevant information based on semantic similarity rather than keyword matching.</p><p>Figure 4: Architecture of an Advanced RAG system with dense embeddings</p><p>Advanced RAG introduces improvements like:</p><ul><li><strong>Neural Re-ranking:</strong> Further refining retrieved documents based on relevance</li><li><strong>Iterative Retrieval:</strong> Multi-step retrieval for complex queries (like multi-hop reasoning)</li></ul><p><strong>Example:</strong> For the query “What treatments work best for seasonal depression?”, the system can understand this is related to “Seasonal Affective Disorder” and retrieve relevant medical literature.</p><h3>2.3.3 Modular RAG</h3><p>Modular RAG combines different retrieval strategies, embedding models, and processing pipelines based on query type and domain.</p><p>A key feature is hybrid retrieval strategies combining:</p><ul><li><strong>Sparse retrieval (BM25)</strong> for exact keyword matching</li><li><strong>Dense retrieval (DPR)</strong> for semantic understanding</li></ul><p>Modular RAG can incorporate specialized tools, APIs, and domain-specific retrievers for different knowledge domains.</p><p><strong>Example:</strong> A financial advisory system might use different retrievers for market data, financial regulations, and tax codes, then integrate results from multiple sources.</p><h3>2.3.4 Graph RAG</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*k5FPpwxSuGHtZxnK2mJcsQ.png" /><figcaption>Knowledge representation in a Graph RAG system</figcaption></figure><p>Graph RAG represents knowledge as interconnected entities and relationships, enabling complex reasoning and multi-hop information retrieval.</p><p>Benefits include:</p><ul><li>Improved handling of relational data and causal relationships</li><li>Support for multi-hop reasoning and connecting disparate information</li></ul><p><strong>Example:</strong> “How did the 2008 financial crisis affect housing policies in European countries?” requires connecting financial events to policy changes across different regions and time periods.</p><h3>2.3.5 Agentic RAG</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*zSaXTDuGZvJzlzKY.png" /><figcaption>Example : source —www.<a href="https://superagi.com/wp-content/uploads/2023/10/Raga-retrieval-augmented-generation-and-actions.png">superagi.com</a></figcaption></figure><p>Agentic RAG represents the current frontier, defined as “RAG systems with agent-like capabilities for autonomous operation.”</p><p>These systems can:</p><ul><li>Plan complex retrieval strategies (deciding what to search and how)</li><li>Evaluate and refine their own outputs</li><li>Integrate diverse knowledge sources adaptively</li></ul><p><strong>Example:</strong> Step 1: Identify sub-questions from a complex query, Step 2: Search appropriate sources for each sub-question, Step 3: Synthesize information into a coherent response</p><h3>2.4 RAG System Challenges</h3><p>ChallengeDescriptionExample<strong>Relevance</strong>Retrieving truly relevant information”What’s the best treatment for my condition?” (too vague)<strong>Synthesis</strong>Combining multiple information sources”Compare renewable energy policies across Scandinavian countries”<strong>Technical</strong>Query formulation, ranking algorithmsConverting natural language to effective search queries</p><h3>3. Agentic RAG: Core Principles and Background</h3><p>Agentic RAG combines the strengths of retrieval-augmented generation with agent-like capabilities for reasoning, planning, tool use, and collaboration.</p><p>While LLMs excel at language understanding and generation, they typically lack the ability to:</p><ul><li>Execute multi-step reasoning processes</li><li>Plan complex tasks with interdependencies</li><li>Use external tools effectively</li><li>Evaluate and refine their own outputs</li></ul><p>Agentic RAG addresses these limitations by asking:</p><ul><li>“How can I improve my response?” (Reflection)</li><li>“What steps should I take to solve this problem?” (Planning)</li><li>“What tools do I need to complete this task?” (Tool Use)</li><li>“How can multiple specialists collaborate on this?” (Multi-Agent Collaboration)</li></ul><p>ComponentDescription</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_9oRc7t0vbasX0CcuhRfTw.png" /></figure><ul><li><strong>LLM Core : </strong>Provides language understanding and generation capabilities</li><li><strong>Memory Systems : </strong>Both short-term (session) and long-term (persistent storage) memory</li><li><strong>Self-Evaluation : </strong>Ability to critique and refine its own outputs (Reflection &amp; Self-Critique)</li><li><strong>Tool Integration</strong>Connection to databases, search engines, APIs, and specialized services</li></ul><p>Figure 6: Core components of an Agentic RAG system</p><h3>3.1 Agentic Design Patterns (The Four Pillars)</h3><p>Figure 7: The four pillars of Agentic RAG design</p><h4>3.1.1 Reflection</h4><p>Reflection enables agents to evaluate their own outputs and reasoning processes, asking: “How can I improve my response?”</p><p>Key implementation approaches include:</p><ul><li><strong>CRITIC</strong>: Dedicated evaluation components</li><li><strong>Reflexion</strong>: Learning from past interactions</li><li><strong>Self-Refine</strong>: Iterative improvement of outputs</li></ul><p>Reflection creates a self-improving cycle:</p><ul><li>Phase A: Generate initial response</li><li>Phase B: Evaluate response for accuracy/completeness</li></ul><p><strong>Example:</strong> After responding to a complex medical query, the system might note: “I should verify the drug interaction information and add dosage guidelines to make this response more complete and precise.”</p><h4>3.1.2 Planning</h4><p>Planning enables breaking down complex tasks into manageable steps, asking: “What’s the best approach to solve this problem?”</p><p>Approaches range from simple to complex:</p><ul><li>Task decomposition into subtasks</li><li>Programmatic planning of execution steps</li><li>Single LLM program with chain-of-thought reasoning</li></ul><p><strong>Example:</strong> “To compare renewable energy policies, I’ll first: 1) Retrieve data on each Scandinavian country’s current policies, 2) Identify key metrics for comparison, 3) Assess implementation timelines, 4) Analyze effectiveness data, 5) Synthesize findings into a comparative analysis.”</p><h4>3.1.3 Tool Use</h4><p>Tool use capabilities allow agents to interact with external systems, databases, SQL engines, and APIs.</p><p>Implementations include:</p><ul><li>GPT-4 Function Calling</li><li>LangChain Tool integration</li><li>Custom API connectors</li></ul><p><strong>Example:</strong> “To answer this question about current market trends, I need to query the financial database with this SQL statement, then call the stock price API to get real-time data.”</p><h4>3.1.4 Multi-Agent Collaboration</h4><p>Multi-agent collaboration involves specialized agents (e.g., researcher, critic, writer) working together on complex tasks.</p><p>This approach brings:</p><ul><li>Scalability through division of labor</li><li>Specialization for complex domains</li><li>Adaptability to changing requirements</li></ul><p>Frameworks like AutoGen, Crew AI, and LangGraph enable multi-agent systems.</p><p><strong>Example:</strong> Agent A researches scientific literature, Agent B analyzes statistical data, Agent C evaluates medical implications, Agent D synthesizes a comprehensive report on a new treatment approach.</p><p>PatternKey Benefit</p><ul><li><strong>Self-Critique</strong>Improved accuracy through continuous refinement</li><li><strong>Prompt Chaining</strong>Complex workflows through sequential processing</li><li><strong>Voting/Parallelization</strong>Enhanced reliability through consensus</li><li><strong>Orchestrator-Worker</strong>Efficient task distribution and coordination</li></ul><h3>4. Agentic Workflow Patterns</h3><p>Agentic Workflow Patterns define the strategies for how agents organize their work processes — structuring information flow, decision-making, and task execution.</p><h3>4.1 Prompt Chaining</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bN5XReVZjWCJGyQl5iJQyQ.png" /></figure><p>Figure 8: Sequential processing in a Prompt Chaining workflow</p><p>Prompt Chaining passes outputs from one processing step as inputs to the next, creating a sequential workflow where each step builds on previous results.</p><p>Benefits include:</p><ul><li>Breaking complex tasks into manageable steps</li><li>Maintaining context across processing stages</li></ul><h3>4.2 Routing</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_a-C4enfebLbr1ZbW4UJkQ.png" /></figure><p>Figure 9: Decision-based routing in an Agentic system</p><p>Routing directs queries to the most appropriate specialized components based on the query type and requirements.</p><p>Applications include:</p><ul><li>Choosing between medical, legal, or financial knowledge bases</li><li>Selecting between GPT-3.5 (faster) vs. GPT-4 (more powerful) based on query complexity</li><li>Routing to specialized retrievers or generation models</li></ul><h3>4.3 Parallelization</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kD1YJdy4CU9ILNWrEg-qXg.png" /></figure><p>Figure 10: Parallel processing in an Agentic RAG system</p><p>Parallelization executes multiple processes simultaneously for efficiency and enhanced reliability:</p><ul><li><strong>Sectioning:</strong> Dividing documents or tasks among multiple agents</li><li><strong>Voting:</strong> Using multiple agents to cross-check results and build consensus</li></ul><h3>4.4 Orchestrator-Workers</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bw2emVdCEJB7zrsBgBGb6w.png" /></figure><p>Figure 11: Hierarchical structure in an Orchestrator-Workers pattern</p><p>This pattern uses a hierarchical structure where a central orchestrator agent coordinates specialized worker agents:</p><ul><li>Orchestrator: Plans strategy, assigns tasks, integrates results</li><li>Workers: Execute specialized tasks (research, analysis, writing)</li></ul><p>Benefits include:</p><ul><li>Clear division of responsibilities</li><li>Combining planning + execution + integration</li><li>Scalability for complex tasks</li></ul><h3>4.5 Evaluator-Optimizer</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FtykHPdNgHunoEL3L-bOQg.png" /></figure><p>Figure 12: Feedback loop in an Evaluator-Optimizer pattern</p><p>This pattern implements continuous improvement through dedicated evaluation and optimization components:</p><ul><li>Evaluator: Assesses outputs for accuracy, completeness, and quality</li><li>Optimizer: Improves processes based on evaluation feedback</li></ul><p>Applications include:</p><ul><li>Fact-checking and source verification</li><li>Output quality enhancement</li></ul><p>Workflow PatternStrengthsWeaknessesBest ForPrompt ChainingSimple implementation, clear sequenceError propagation, limited parallelismSequential tasks with clear dependenciesRoutingEfficient resource use, specializationComplex decision logic, maintenanceSystems handling diverse query typesParallelizationEfficiency, redundancy for reliabilityResource intensive, coordination overheadTime-sensitive tasks, consensus buildingOrchestrator-WorkerClear hierarchy, specialized executionOrchestrator bottlenecks, complexityComplex projects requiring diverse expertiseEvaluator-OptimizerContinuous improvement, quality controlAdded latency, resource requirementsHigh-stakes applications requiring accuracy</p><h3>5. Agentic RAG Architectures</h3><p>Agentic RAG architectures define the structural organization of agent components, their interactions, and information flow patterns.</p><p>ArchitectureComponentsInteractionsInformation FlowApplications<strong>Single-agent</strong>One LLM with multiple capabilitiesSelf-reflection, tool useLinear processingGeneral-purpose assistants<strong>Multi-agent</strong>Multiple specialized agentsCollaboration, negotiationComplex, multi-directionalResearch, complex analysis<strong>Agent-Tool</strong>Agents with integrated toolsTool selection, data processingAgent-to-tool requestsData-intensive Q&amp;A<strong>Agent-Human</strong>Agents collaborating with humansGuidance seeking, explanationInteractive dialogueCreative tasks, education</p><h3>5.1 Single-agent Architectures</h3><p>Single-agent architectures use one LLM with multiple capabilities to handle the entire task workflow — from planning to execution to evaluation.</p><p>These systems employ:</p><ul><li>Self-reflection mechanisms</li><li>Comprehensive tool use</li><li>Internal planning processes</li></ul><p>The main advantage is simplicity and coherence across the workflow.</p><h3>5.2 Multi-agent Architectures</h3><p>Multi-agent architectures distribute tasks among specialized agents that collaborate to solve complex problems:</p><ul><li>Agent A: Research specialist finding relevant information</li><li>Agent B: Critical thinking specialist evaluating information</li><li>Agent C: Communication specialist producing final output</li></ul><p>These agents communicate and coordinate to leverage collective capabilities.</p><p>The main advantage is specialization and the ability to handle complex, multi-faceted tasks.</p><h3>5.3 Agent-Tool Architectures</h3><p>Agent-Tool architectures focus on integrating LLM agents with external tools and services:</p><ul><li>Vector databases (FAISS, Pinecone)</li><li>External APIs</li><li>Specialized processors: calculators, SQL engines, code interpreters</li></ul><p>The agent serves as an intelligent coordinator, selecting and using appropriate tools as needed.</p><p>The main advantage is expanded capabilities through specialized external resources.</p><h3>5.4 Agent-Human Architectures</h3><p>Agent-Human architectures explicitly incorporate human collaboration in the workflow:</p><ul><li>Agents request clarification or verification from humans</li><li>Humans provide guidance, feedback, or specialized knowledge</li><li>The system balances automation with human expertise</li></ul><p>The main advantage is combining AI efficiency with human judgment and domain expertise.</p><h3>6. Applications of Agentic RAG</h3><p>Agentic RAG systems are finding applications across diverse domains, leveraging their advanced capabilities for complex information processing and problem-solving.</p><h3>6.1 Education &amp; Tutoring</h3><p>In educational contexts, Agentic RAG provides personalized learning experiences, adaptive explanations, and comprehensive tutoring:</p><ul><li><strong>Capabilities:</strong> Curriculum-aligned knowledge retrieval, adaptive difficulty adjustment</li><li><strong>Benefits:</strong> Personalized learning paths, detailed explanations tailored to student knowledge</li></ul><p>Agentic patterns commonly used:</p><ul><li><strong>Prompt Chaining</strong> + <strong>Evaluator-Optimizer</strong> for adaptive instruction</li></ul><h3>6.2 Document Summarization &amp; Analysis</h3><p>For document processing, Agentic RAG enables sophisticated analysis of large document collections:</p><ul><li><strong>Process:</strong> Agent A extracts key information, Agent B identifies trends, Agent C synthesizes findings</li><li><strong>Benefits:</strong> Comprehensive analysis, focus on important patterns, contextual understanding</li></ul><p>Agentic patterns commonly used:</p><ul><li><strong>Multi-Agent</strong> + <strong>Evaluator</strong> for collaborative analysis</li></ul><h3>6.3 Complex QA &amp; Research</h3><p>For advanced research tasks requiring deep investigation:</p><ul><li><strong>Example:</strong> “What are the emerging AI applications in healthcare diagnostics?”</li></ul><p>Agentic patterns commonly used:</p><ul><li><strong>Single-Agent</strong> + <strong>Tool Use</strong> for comprehensive research</li><li><strong>Routing</strong> for accessing specialized knowledge sources</li></ul><h3>6.4 Legal/Policy Reasoning</h3><p>For legal analysis requiring precise interpretation and application of complex regulations:</p><p>Agentic patterns commonly used:</p><ul><li><strong>Prompt Chaining</strong> + <strong>Agent-Human Collaboration</strong> for nuanced legal reasoning</li></ul><h3>6.5 Business Intelligence</h3><p>For business analysis combining diverse data sources and analytical techniques:</p><p>Agentic patterns commonly used:</p><ul><li><strong>Agent-Tool</strong> + <strong>Planner-Worker</strong> for data-driven business insights</li></ul><h3>6.6 Scientific Discovery</h3><p>For scientific research requiring hypothesis generation and testing:</p><p>Agentic patterns commonly used:</p><ul><li><strong>Orchestrator-Worker</strong> + <strong>Evaluator</strong> for rigorous scientific methodology</li></ul><h3>7. Challenges and Future Directions for Agentic RAG</h3><p>While Agentic RAG offers powerful capabilities, several challenges and future research directions remain.</p><h3>7.1 Planning &amp; Reasoning</h3><p>Current limitations in agent planning and reasoning capabilities include:</p><ul><li>Difficulty with very complex, multi-step planning</li><li>Limited causal reasoning</li><li>Challenges with task prioritization</li></ul><p>Potential solutions include:</p><ul><li>Enhanced LLM reasoning abilities</li><li>Improved context management</li><li>Advanced Chain-of-Thought prompting techniques</li></ul><h3>7.2 Evaluation</h3><p>Evaluating Agentic RAG systems remains challenging:</p><ul><li>How do we measure the quality of agent reasoning?</li><li>How do we assess response completeness and accuracy?</li></ul><p>Emerging approaches include:</p><ul><li>LLM-based evaluation frameworks (Self-Evaluation)</li><li>Task-specific metrics for different applications</li></ul><h3>7.3 Performance Optimization</h3><p>Performance bottlenecks include:</p><ul><li>Latency in multi-step agent workflows</li><li>API rate limitations</li><li>GPU and computational resource requirements</li></ul><p>Future directions include:</p><ul><li>Adaptive workflows that balance speed and quality</li><li>More efficient retrieval and processing pipelines</li></ul><h3>7.4 Security and Access Control</h3><p>Security challenges include:</p><ul><li>Safe access to external resources and APIs</li><li>Information leakage prevention</li></ul><p>Future directions include:</p><ul><li>Better API authentication mechanisms</li><li>Agent governance frameworks</li></ul><h3>7.5 Memory Management</h3><p>Long-term memory challenges include:</p><ul><li>Managing agent context over extended interactions</li><li>Balancing memory retention with processing efficiency</li></ul><p>Future directions include:</p><ul><li>Hierarchical memory architectures</li><li>Retrieval-based memory systems</li></ul><h3>Conclusion</h3><p>Agentic RAG represents a significant advancement in AI systems, combining the knowledge retrieval capabilities of traditional RAG with agent-like abilities for reasoning, planning, tool use, and collaboration.</p><p>As these systems continue to evolve, we can expect:</p><ul><li>More sophisticated reasoning and planning capabilities</li><li>Enhanced multi-agent collaboration frameworks</li><li>Broader integration with specialized tools and APIs</li><li>More efficient performance in complex tasks</li></ul><p>The future of AI assistance likely lies in these agent-augmented systems that can not only retrieve relevant information but also reason about it, plan complex actions, and collaborate with humans and other agents to solve increasingly sophisticated problems.</p><p>Agentic RAG is still in its early stages, but its potential to transform how we interact with AI systems and leverage their capabilities is immense. As researchers and developers continue to address current challenges, we can look forward to even more capable and useful AI assistants in the coming years.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b9317f09eb06" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How can we empower girls in African continent with App-development education?]]></title>
            <link>https://medium.com/@hailey_dataforsocialgood/how-can-we-empower-girls-in-african-continent-with-app-development-education-b4d7dd45f658?source=rss-b75ccbbcad05------2</link>
            <guid isPermaLink="false">https://medium.com/p/b4d7dd45f658</guid>
            <category><![CDATA[responsible-ai]]></category>
            <category><![CDATA[edtech]]></category>
            <category><![CDATA[tech-for-good]]></category>
            <dc:creator><![CDATA[Hailey Jun]]></dc:creator>
            <pubDate>Fri, 08 Nov 2024 08:51:52 GMT</pubDate>
            <atom:updated>2024-11-08T08:52:05.365Z</atom:updated>
            <content:encoded><![CDATA[<h3>How can we empower girls with App-development education?</h3><p>Review of Google.Org impact Summit 2024 — App Inventor Foundation</p><p><a href="https://www.appinventorfoundation.org/">App Inventor Foundation</a></p><p>App inventor foundation is founded by educators and researchers at MIT and Google to enable everyone to create mobile apps that can change their community.</p><p>CEO &amp; Executive Director of App Inventor foundation, <a href="https://www.linkedin.com/in/natalie-lao/">Dr.Natalie Lao</a> gave speech at Google.Org Impact Summit, introducting cases where girls made impact for their communities.</p><h4><strong>1. App for water quality assurance.</strong></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*C4mjtUI1Tm6uNptqU6KXuQ.png" /></figure><p>Gitanjali Rao, at age of 15, developed an IoT enabled bluetooth sensor that hooks up to cell phone and light when there is problem with the drinking water. Her latest app is an app called kindely, which helps cyberbullying problem with sentiment analysis.</p><h4>2. Hackathon for Good</h4><p>App Inventor Foundation launched global AI hackathon to get as many people as possible to make them create app for good. The median age of participants were 16 and 337 girls and women participated.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lzjE-IEXdn6TiL9NZwNAeQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*X7m9kyYYNasqXnNe0dhybA.png" /></figure><p>The winner of the hackathon — Maura Moore from Ireland — created an app for visually impaired people. She kept on working with Vision Island (National sigh-loss agency of Ireland) to refine and scale the app to make real impact.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b4d7dd45f658" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Journey from UN Intern to a data scientist — 2. Why I sold coffee to build toilet in Laos]]></title>
            <link>https://medium.com/@hailey_dataforsocialgood/my-career-journey-starting-at-humanitarian-sector-to-land-at-business-leading-to-data-for-social-a1843368928b?source=rss-b75ccbbcad05------2</link>
            <guid isPermaLink="false">https://medium.com/p/a1843368928b</guid>
            <dc:creator><![CDATA[Hailey Jun]]></dc:creator>
            <pubDate>Sun, 15 Oct 2023 12:51:30 GMT</pubDate>
            <atom:updated>2023-10-15T13:43:26.699Z</atom:updated>
            <content:encoded><![CDATA[<p><em>My career journey starting at humanitarian sector to land at business, leading to data for social good.</em></p><p>For the intriguing beginning of my journey, see Chapter 1, where I dive into my initial career steps and learnings from a UN internship 👉 <a href="https://medium.com/@hailey_dataforsocialgood/embarking-on-a-career-journey-my-six-month-internship-at-the-un-77d64c4cf460">Read here</a></p><h3>Chapter 2. Selling Tumblers for Cleaner Toilet?</h3><p>After I came back from UN internship at Bangkok, I got re-immersed to my usual university days. Returning to my student NGO and entering my final year of university, I was engulfed in a passion for alleviating global poverty. This seemingly naive fixation became the foundation that led me, boundlessly, into the world of business.</p><h3>Problems ahead</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*t7YKJaGc7LBPJXUe9XRseA.jpeg" /><figcaption>Myself at the senior year of university — elected as the leader of a student NGO</figcaption></figure><p>Shortly after I stepped into the leadership role of the student NGO, a daunting issue emerged: our funding disappeared. With contracts for toilet construction at a community center in Lao PDR already in place, we were perilously close to losing everything.</p><h3><strong>Innovative Approaches?</strong></h3><p>Instead of looking for new funding, I was determined to generate our own revenue. Our plan involved <strong><em>selling tumblers and fair-trade coffee, thereby raising funds while also advocating for sustainability.</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LbEXF9SNm4lfurQKmlrcnw.jpeg" /></figure><p>Our approach involved:</p><ul><li>Securing initial funding</li><li>Establishing our financial goal</li><li>Designing and sourcing the product</li><li>Developing a sales strategy, including channel development</li><li>Utilizing social media for marketing and sales promotion</li></ul><h3><strong>Our Strategy in Action</strong></h3><ul><li>We secured seed money through UNESCO Korea’s youth support program and a commercial bank in Korea.</li><li>Identified the necessary fundraising target.</li><li>Attentively designed and sourced the products, including:<br>1) Curating and creating the coffee menu<br>2) Designing the tumbler, featuring a beautifully crafted logo</li><li>Ensuring product quality and cost-effectiveness through price comparisons</li><li>We chose to operate at a university festival, maximizing customer exposure.</li><li>Crafted marketing materials and ran our advertisements, while also exploring additional sales channels to prevent inventory stagnation.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*c-1jKYbdLDVV5bkjGhxuyQ.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XvFm-oeMg0GuPeDQNiTsrw.jpeg" /><figcaption>Left ) We also had to wash the tumblers ourselves before the sales. Right) The poster we displayed for offline marketing.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*F6TKtktj4LDPCKzzVqaJMA.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8ikVUDzc5okuER2hnjk3pA.jpeg" /><figcaption>Our sales booth</figcaption></figure><h3>The results</h3><p>So, finally we successfully raised the target amount of money. Although the school festival sales were not enough, we further accelerated sales through additional sales channels.</p><p>Few months after, we received such photos with report of the money spent.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jgO8qiy3lhrngmUQn2tz9A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ou08aAm2YDhayBCu8daRWQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*E9OzjiK3DElB8lmt1t4Lqw.png" /><figcaption>The toilet construction afterwards. The profits raised were fully donated to construction of toilet construction and also other relief activities in Laos.</figcaption></figure><h3><strong>Key Learnings</strong></h3><p>In a few hectic months, amidst the excitement and joy of project execution, my learnings were profound:</p><ol><li><strong>The Complexity of Small Business<br></strong>The extensive to-do list and strategic execution demonstrated that even a small business can be multifaceted and requires dedicated, often behind-the-scenes, effort to succeed.</li><li><strong>The Selectivity of Heeding Advice<br></strong>Advice is a blend of wisdom and misdirection. Discernment is crucial. Despite facing criticism and negativity, the shared passion among my colleagues propelled us forward.</li><li><strong>The Empowerment of Small Wins<br></strong>Reflecting on my university years, this small business endeavor stands out. Not only did it instill a powerful “can-do” attitude in me, but it also paved the way towards joining start-ups and exploring the business world.</li></ol><p><strong>What’s Next?</strong></p><p>This all felt like the old story of selling orange juice at a business school. However, I was so excited for what we have achieved.</p><p>Witnessing the robust, dynamic world of business, akin to an ever-churning machine generating finances, was captivating. Just as selling orange juice can unveil business nuances, I discovered a passion for unraveling business challenges.</p><p>I saw the potential to extract learnings from business and apply them to humanitarianism. The diverse nature of the two sectors hinted at learnings that could be mutually beneficial.</p><p>Therefore, the next chapter of my journey ventures deeper into the business world.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a1843368928b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Journey from UN Intern to a data scientist — 1. What I learned in UN 6 years ago]]></title>
            <link>https://medium.com/@hailey_dataforsocialgood/embarking-on-a-career-journey-my-six-month-internship-at-the-un-77d64c4cf460?source=rss-b75ccbbcad05------2</link>
            <guid isPermaLink="false">https://medium.com/p/77d64c4cf460</guid>
            <category><![CDATA[humanitarian]]></category>
            <category><![CDATA[international-development]]></category>
            <category><![CDATA[united-nations]]></category>
            <dc:creator><![CDATA[Hailey Jun]]></dc:creator>
            <pubDate>Sun, 01 Oct 2023 06:37:17 GMT</pubDate>
            <atom:updated>2023-10-01T08:10:54.648Z</atom:updated>
            <content:encoded><![CDATA[<p><em>My career journey starting at humanitarian sector to land at business, leading to data for social good.</em></p><p>Chatper1</p><p>Since my teenage years, the pervasive poverty in some countries has piqued my interest and stirred questions within me. “Why is there so much poverty in some countries? Why do some individuals have to endure suffering while others remain indifferent?” My quest for answers led me through numerous books, particularly focusing on the poverty in the African continent, yet the questions lingered until my twenties.</p><p>After a few years, I entered university. As a young, empathetic woman with a mission, I was fortunate enough to secure an internship at the United Nations (UN) after my third year of university. The journey took me from my home in South Korea to Bangkok, the Asia-Pacific regional hub of the UN, and it was here that a six-month adventure that would profoundly shape my perspectives commenced.</p><h3><strong>What I did as UN intern</strong></h3><p>My team was focused on economic data analytics. My day-to-day work included the following :</p><ul><li>10% — Revising a policy brief (A policy suggestion to the government officials)</li><li>60% — Project development. This included : Reading project references, designing a project tailored to our team, creating slides and presenting to relevant stakeholders</li><li>20% — Data analytics with simple visualizations (Downloading statistical dataset such as one related to global financial flow and making the data easy to understand)</li><li>10% — Other tasks related to learning, sharing knowledges and team building activities.</li></ul><p>I was only 23 and it was my first experience in a workplace, outside of school. It was my first time to actually use excel or spreadsheet.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qmNJCTDQJnqq-8LV66RlIw.jpeg" /><figcaption>My desk panel filled with memos</figcaption></figure><p><strong>My team</strong></p><p>I was lucky enough to meet a wonderful supervisor and a collegue to be my role model and provide amazing guidance. My supervisor, was so kind to give me a detailed guidance and explanation, whenever I asked questions. My colleague did an excellent job at work in a proactive manners. Until now, she gives me lots of inspiration</p><h3>What I learned during UN internship?</h3><ol><li><strong>Read, Read, READ!!</strong></li></ol><p>Confidently, I can say that my journey was deeply intertwined with extensive reading. Given that the bulk of my work was centered around project development, reading voluminous documents became a routine.</p><ul><li><strong>Typically, a project paper spanned over 40 dense pages.</strong></li><li><strong>I read more than 10 paper for one project development.</strong></li><li><strong>While I cannot recall the exact volume, there were weeks where my sole task was to meticulously peruse through all related projects.</strong></li></ul><p>My love for reading transformed this into an enjoyable endeavor. In retrospect, I believe it was this rigorous engagement with texts that enabled the contributions of university students to be integrated into expansive international work.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NMpsaUBnORu-lbBo5BKxZQ.jpeg" /><figcaption>I used to read project papers as benchmark and create project plan for our team</figcaption></figure><p><strong>2. Self-objectification : Viewing myself in third person’s point of view</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3bALXjiyDs8KqBLD0hNLsg.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*l1iW_MSXxko68UhBBmNlTQ.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Q6l64a-EkcQ3-HXc-vtv-w.jpeg" /><figcaption>My first impression about Bangkok, Thailand in 2017</figcaption></figure><p><strong><em>The pivotal lesson I learned was the realization of my own narrow-mindedness.</em></strong><em> </em>Bangkok and UN was a melting pot of diversity, introducing me to people from origins I’d barely heard of. Encountering individuals from such varied backgrounds opened my eyes to a world vastly different from my own. It enlightened me about my standardized notion of success, and how I was suppressing myself to follow the path.</p><p><strong>3. Career path afterwards</strong></p><p><strong><em>The second profound lesson was recognizing that the UN might not be the optimal starting point for my career.</em></strong></p><ul><li>UN requires at least Masters degree most of the times</li><li>Being a specialist / having diverse experience could be a benefit in terms of career development</li><li>UN is not the only choice. You can learn elsewhere and come back here.</li></ul><p>Listening through many advises regarding career at UN, I came to a conclusion that remaining at the UN might not foster my growth in the direction I aspired towards, especially given my awareness of my limited knowledge in many areas. For this, I thought I needed my experience outside UN and <strong>comeback here sometimes later.</strong></p><p><strong>(+ This is the first time I truly realized the importance of reaching out to seniors for advice)</strong></p><p><strong>A Continuous Journey of Learning and Growth</strong></p><p>Upon returning from the UN, I didn’t immediately prepare to pursue a career there. Instead, I sought alternative paths that would enhance my learning curve and broaden my horizons. That includes business, strategy and data. Now, six years post-internship, my journey continues to unfold.</p><p>✅Please follow on linked <a href="https://www.linkedin.com/in/halim-jun-1b09b9155/">https://www.linkedin.com/in/halim-jun-1b09b9155/</a> and Medium if you want to hear more</p><p>✅(*I tried to keep the contents as general as possible. Contact if any part of this article should be deleted)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=77d64c4cf460" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How is USAID Driving Lifesaving Responses with Data? — FEWS NET]]></title>
            <link>https://medium.com/@hailey_dataforsocialgood/how-is-usaid-driving-lifesaving-responses-with-data-fews-net-e1aad60647fd?source=rss-b75ccbbcad05------2</link>
            <guid isPermaLink="false">https://medium.com/p/e1aad60647fd</guid>
            <category><![CDATA[social-innovation]]></category>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[usaid]]></category>
            <category><![CDATA[data-for-good]]></category>
            <dc:creator><![CDATA[Hailey Jun]]></dc:creator>
            <pubDate>Sun, 18 Jun 2023 08:08:41 GMT</pubDate>
            <atom:updated>2023-06-18T08:08:41.609Z</atom:updated>
            <content:encoded><![CDATA[<h3>How is USAID Driving Lifesaving Responses with Data? — FEWS NET</h3><p>Early warning system of USAID. Referencing the presentation by FEWS NET at <a href="https://www.digitaldevforum.com/">https://www.digitaldevforum.com/</a></p><p><a href="https://fews.net/">HOME | FEWS NET</a></p><h3>What Is Fews Net</h3><p>The Famine Early Warning Systems Network (FEWS NET) is a significant provider of early warning and analysis regarding acute food insecurity worldwide, established by USAID.</p><p>Its objective is to anticipate acute food insecurity before it occurs by utilizing various datasets, including climate/weather and food prices. <strong>By forecasting food crises and identifying specific regions, FEWS NET aims to allocate more financial resources (of USAID) in advance, thereby</strong> preventing such disastrous situations and saving thousands of lives.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_qx0XOTo9kye2L8mEOtTNQ.png" /><figcaption>Near Term Acute Food Insecurity Classification by FEWS NET</figcaption></figure><p>FEWS NET’s approach involves monitoring and <strong>forecasting acute food insecurity globally</strong> [<a href="https://fews.net/">2</a>]. Their analysis evaluates the severity of food insecurity by assessing households’ ability to meet basic survival needs and maintain normal livelihoods.<strong> They classify the likely outcomes of acute food insecurity in the medium-term (8 months into the future)</strong> projection periods [<a href="https://fews.net/">2</a>]. This classification system helps identify areas that require emergency response and aid.</p><h3>How Do Fews Net Team work</h3><p>To achieve accurate predictions, FEWS NET <strong>considers various factors that contribute to acute food insecurity, such as climate, conflict, and markets/trade</strong> [<a href="https://fews.net/">2</a>]. They have an Agroclimatology Team that utilizes remote sensing satellite imagery to monitor and forecast climatic conditions in the most food-insecure countries [<a href="https://fews.net/">2</a>]. Additionally, they gather information and develop assumptions about conflict and its impact on food insecurity, while also monitoring market prices, staple food production, macroeconomic indicators, trade policies, and market shocks [<a href="https://fews.net/">2</a>].</p><h3>USE CASES</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*q8xR8_DZupdVzVVI4fXnVg.png" /></figure><p>For example, last 2022, the Food insecurity was alerted for Ethiopia. This led to increase preemptive allocation of funds from USAID for the region.</p><h3>Accuracy and Performance of FEWS NET forecast</h3><p>Research has shown that FEWS NET’s projections of food security in Africa have an <strong>overall accuracy of 84 percent</strong>, although the accuracy drops when it tends to over-project severe food insecurity [<a href="https://www.sciencedirect.com/science/article/pii/S2211912421000201">3</a>]. The study suggests that humanitarian assistance, unanticipated climate events, and conflict shocks can affect the accuracy of projections [<a href="https://www.sciencedirect.com/science/article/pii/S2211912421000201">3</a>]. Therefore, ongoing validation, deeper scrutiny of reliability factors, increased transparency, and informed usage are recommended to improve FEWS NET’s performance [<a href="https://www.sciencedirect.com/science/article/pii/S2211912421000201">3</a>].</p><h3><strong>How does the FEWS NET Tool look like ?</strong></h3><ul><li>Data warehouse is set up, and by 2023, 19,250,972 data points are stored</li><li>Using APIs, various dataset is aggregated into one storage. Such data include climatology, agriculture, livelihoods, nutrition, conflict, floods, market/trade/price and infectious diseases.</li></ul><p>Simply speaking, <strong>the pipeline</strong> can be illustrated as below. Different data sources are stored in FEWS NET data warehouse. From the data warehouse, the data is utilized by other parties.</p><ol><li><strong>Fews Net Data pipeline</strong></li></ol><figure><img alt="Data pipieline" src="https://cdn-images-1.medium.com/max/1024/1*6lZRHgIaVgPv4n8hNg64cA.png" /><figcaption>FEWS NET data pipeline</figcaption></figure><p>*Kobo : data collection tool</p><p><strong>2. FEWS data explorer</strong></p><p>It is open provider of various dataset where one can easily access the necessary information to predict food insecurity.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*k4Ihcwqbd_Vk--NkDaBnHQ.png" /></figure><h3>Thoughts</h3><ol><li>The value of the FEWS NET early warning system lies in its ability to make preemptive decisions, which reduces costs and saves lives compared to reactive measures. The <strong>promptness</strong> of the alert is of utmost importance since the faster it can predict a food crisis, the more time there is to take appropriate measures. By allocating resources and implementing preventive measures in a timely manner, the impact of the crisis can be mitigated, leading to a significant reduction in the loss of lives [<a href="https://documents.worldbank.org/en/publication/documents-reports/documentdetail/304451600783424495/predicting-food-crises">4</a>].</li><li><strong>Accuracy</strong> is another crucial factor that decision-makers seek in an early warning system. A high level of accuracy ensures that fund allocation decisions are efficient, avoiding the wastage of resources. By providing reliable and precise information, the system enables decision-makers to respond effectively to food crises. This accuracy is essential for optimizing resource allocation and maximizing the number of lives saved globally [<a href="https://www.sciencedirect.com/science/article/pii/S0048969721024372">3</a>].</li><li>Furthermore, the remarkable aspect of FEWS NET is its ability to <strong>pinpoint the specific regions</strong> within a country that are vulnerable to famine. This <strong>targeted</strong> <strong>approach</strong> enables the efficient distribution of aid, resulting in optimal fund allocation and the potential to save more lives. By identifying the most affected areas, resources can be directed to where they are needed the most, ensuring a more effective response to food insecurity crises worldwide.</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e1aad60647fd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Uber For Tractor in Africa — Hello Tractor]]></title>
            <link>https://medium.com/@hailey_dataforsocialgood/uber-for-tractor-in-africa-hello-tractor-ce48a7a97ef3?source=rss-b75ccbbcad05------2</link>
            <guid isPermaLink="false">https://medium.com/p/ce48a7a97ef3</guid>
            <category><![CDATA[data-for-good]]></category>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[sustainability]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[tech-for-good]]></category>
            <dc:creator><![CDATA[Hailey Jun]]></dc:creator>
            <pubDate>Sun, 30 Apr 2023 15:36:35 GMT</pubDate>
            <atom:updated>2023-04-30T15:55:26.337Z</atom:updated>
            <content:encoded><![CDATA[<h3>Uber For Tractor in Africa — Hello Tractor</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*myBjXTVAL9LRfTTKa2p5_g.png" /><figcaption><em>(State Dept./Doug Thompson)</em></figcaption></figure><h3>Story from Global Digital Development Forum</h3><p><a href="https://www.digitaldevforum.com/">Global Digital Development Forum</a></p><p><strong>Hello Tractor is an innovative ag-tech start-up </strong>with a presence in Kenya and Nigeria that is addressing the challenges faced by <strong>smallholder farmers in accessing tractors</strong>. The speaker highlighted that, while agriculture accounts for around 35% of the African economy, compared to 5.4% in the US, <strong>more than 70% of farmers in Africa do not own a tractor due to the small scale of their land ownership.</strong></p><p>This lack of tractor ownership leads to a significant loss of productivity, estimated at around 50%, which has a negative impact on the farmers&#39; income and livelihoods. However, Hello Tractor&#39;s mobile tractor-sharing platform is tackling this issue by providing farmers with access to affordable and convenient tractor services.</p><p>By leveraging mobile technology, <strong>Hello Tractor connects farmers with local tractor owners and operators, enabling them to request and book tractor services on-demand.</strong> This approach has proved to be highly effective, as it allows farmers to increase their productivity and yields while reducing labor costs and time. Furthermore, it provides tractor owners and operators with additional income opportunities and boosts their profitability.</p><p><a href="https://hellotractor.com/">https://hellotractor.com/</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/522/1*DNrCoSTjQH9ZEp38ygubhw.png" /><figcaption>Example of Hello Tractor App Interface</figcaption></figure><ul><li>HelloTractor is a sharing platform that enables farmers to share tractors without owning the machinery themselves.</li><li>The platform allows operators to add themselves to a list, which users can view to find available operators within their location and book them with ease through the app.</li><li>Users can also view nearby tractors available for rent through the platform</li><li>HelloTractor also collects and utilizes various data for monitoring purposes, such as fuel levels and maintenance information. This data enables users to better manage their equipment and optimize their farming operations.</li></ul><h3>My Thoughts</h3><ul><li>The innovative approach of using IoT to improve agricultural productivity is truly impressive. By collecting various mobility data, issues such as fraud and vehicle safety can be addressed, further enhancing the platform’s value to farmers.</li><li>Moreover, <strong>the active user base within the platform </strong>presents opportunities for other businesses to leverage. For instance, there is potential to offer commerce or information-sharing services that cater to local needs and further support the agricultural community. (However, as I am not familiar with the local context, it would be inappropriate for the local’s setting)</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ce48a7a97ef3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Where data is utilized for social good (2) Meta (Facebook)]]></title>
            <link>https://medium.com/@hailey_dataforsocialgood/where-data-is-utilized-for-social-good-2-meta-facebook-6a4a28c35692?source=rss-b75ccbbcad05------2</link>
            <guid isPermaLink="false">https://medium.com/p/6a4a28c35692</guid>
            <category><![CDATA[social-innovation]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[meta]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[international-development]]></category>
            <dc:creator><![CDATA[Hailey Jun]]></dc:creator>
            <pubDate>Wed, 12 Oct 2022 11:25:46 GMT</pubDate>
            <atom:updated>2022-10-15T12:38:38.877Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1014/1*sb1WqUPkT7mncrdfkIfMnQ.png" /><figcaption>Mapping of Meta’s high resolution density map, <a href="https://dataforgood.facebook.com/dfg/tools/high-resolution-population-density-maps">source</a></figcaption></figure><p>While there are various entities that strive to make a positive impact in making the world sustainable with data, the private sector, with its great expertise in data science, is certainly one of them.</p><p>“Data for Good” at meta partners with various organizations including NGOs and governments across the world to make improvements on social issues. Especially, Meta provides various <strong>maps</strong> on population, mobility, infrastructure, economy, and poverty. (You can learn about meta’s data listed<a href="https://dataforgood.facebook.com/dfg/tools"> here</a>)</p><p>Following are some of the featured cases and you can find more in the <a href="https://dataforgood.facebook.com/dfg/about">link</a>.</p><p><a href="https://dataforgood.facebook.com/dfg/about">Facebook Data For Good 简介</a></p><h4>1. Meta’s High Resolution Population Density Estimates to Foster Clean Water in Rwanda and Zambia</h4><p>World Vision has teamed with Meta to bring disruptive change in WASH (Water, Sanitation and Hygiene) In remore area, lack of information on accurate population density of each place is a challenge for ensuring access to clean water.</p><p>Facebook’s Data for Good (D4G) teamed with Columbia University to created a detailed map of the population density using the high-resolution satellite images and machine learning. Specifically, a darkened pixel in the satellite image is where more people are living, thus deep-learning’s pattern recognition seizes such feature and calculates the estimated population density. Combining another map on water collection point with the population map, finally, enabled pinpointing where access to clean water is limited.</p><p>Using such ‘layers of map’, world vision conducted analysis on Zambia and Rwanda to track the shortage of water and listed up primary location for investment where the water shortage is most severe.</p><p>Read more at :</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fstorymaps.arcgis.com%2Fstories%2Fa73563c0d11b433fa35e0bd10a546087&amp;display_name=ArcGIS+StoryMaps&amp;url=https%3A%2F%2Fstorymaps.arcgis.com%2Fstories%2Fa73563c0d11b433fa35e0bd10a546087&amp;image=https%3A%2F%2Fwww.arcgis.com%2Fsharing%2Frest%2Fcontent%2Fitems%2Fa73563c0d11b433fa35e0bd10a546087%2Fresources%2Fpe75u5JubR7LlnexnCS1e.bmp%3Fw%3D800&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=arcgis" width="800" height="600" frameborder="0" scrolling="no"><a href="https://medium.com/media/0f2b23727f2c67cda3bc19ad5e68ebf6/href">https://medium.com/media/0f2b23727f2c67cda3bc19ad5e68ebf6/href</a></iframe><h4>2. Mapping the movement of Ukrainian War refugee</h4><p>Direct Relief, a Non-profit organization based in the US, utilized phone-based mobility data and population density images from Meta, to better understand where refugees are likely to go. Plus, using the social network data from Meta, researchers were able to predict where the refugees have social connections in to find out where the refugees will be heading to.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/828/0*MujYvM401Tdujbuf" /></figure><p>Read more at :</p><p><a href="https://www.directrelief.org/2022/03/where-ukrainian-refugees-are-heading-and-why/">Where Ukrainian Refugees Are Heading - and Why.</a></p><h3>Other relevant data sources for “map”:</h3><ul><li><a href="https://data.humdata.org/">Humanitarian Data Exchange | Find &amp; Use Crisis Data | HDX</a></li><li><a href="https://www.worldpop.org/">Open Spatial Demographic Data and Research - WorldPop</a></li></ul><p><a href="https://www.openstreetmap.org/#map=7/35.948/127.736">https://www.openstreetmap.org</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6a4a28c35692" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>