<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: lochan k m</title>
    <description>The latest articles on DEV Community by lochan k m (@lochan_km_8f2c3f2051e9d1).</description>
    <link>https://dev.to/lochan_km_8f2c3f2051e9d1</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4167339%2F121d8cde-a1c1-42e3-925f-6183ad831474.png</url>
      <title>DEV Community: lochan k m</title>
      <link>https://dev.to/lochan_km_8f2c3f2051e9d1</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/lochan_km_8f2c3f2051e9d1"/>
    <language>en</language>
    <item>
      <title>👻 STRANGER: An AI Agent That Connects Students, Recruiters, Researchers, and Companies with Eachother</title>
      <dc:creator>lochan k m</dc:creator>
      <pubDate>Sat, 31 Oct 2026 11:49:15 +0000</pubDate>
      <link>https://dev.to/lochan_km_8f2c3f2051e9d1/stranger-an-ai-agent-that-connects-students-recruiters-researchers-and-companies-with-753</link>
      <guid>https://dev.to/lochan_km_8f2c3f2051e9d1/stranger-an-ai-agent-that-connects-students-recruiters-researchers-and-companies-with-753</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week4-2026-10-26"&gt;Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h1&gt;
  
  
  👻 STRANGER — The Ghost in the Network That Finds Your Missing Connection
&lt;/h1&gt;

&lt;p&gt;&lt;strong&gt;Everyone is someone's missing connection. What if AI could find it?&lt;/strong&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;In a world with billions of people, finding the right person can still be surprisingly difficult.&lt;/p&gt;

&lt;p&gt;A student needs a hackathon teammate. A researcher needs an assistant. A startup founder needs a technical partner. A recruiter needs the right candidate.&lt;/p&gt;

&lt;p&gt;The people we need might already exist somewhere in our networks, but their skills, interests, and opportunities remain invisible to us.&lt;/p&gt;

&lt;p&gt;That's the problem that inspired &lt;strong&gt;STRANGER&lt;/strong&gt;, an AI-powered human discovery and opportunity-matching agent built using Fetch.ai's Agentverse, ASI:One, and the open-source uAgents framework.&lt;/p&gt;

&lt;h3&gt;
  
  
  How I interpreted "Ghost in the Machine"
&lt;/h3&gt;

&lt;p&gt;I imagined a digital ghost moving through a network of disconnected portfolios, discovering meaningful connections that people might otherwise overlook.&lt;/p&gt;

&lt;p&gt;Instead of haunting machines, STRANGER explores the information people voluntarily share to discover potential collaborators, mentors, researchers, recruiters, and teammates.&lt;/p&gt;

&lt;p&gt;The Halloween theme also inspired the concept of personalized, spooky-themed portfolio illustrations.&lt;/p&gt;

&lt;h3&gt;
  
  
  How STRANGER works
&lt;/h3&gt;

&lt;p&gt;&lt;strong&gt;1. Adaptive onboarding&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;STRANGER is designed to collect meaningful user information through just two or three questions.&lt;/p&gt;

&lt;p&gt;It focuses on:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Skills and experience&lt;/li&gt;
&lt;li&gt;Education and achievements&lt;/li&gt;
&lt;li&gt;Projects and interests&lt;/li&gt;
&lt;li&gt;Goals and career aspirations&lt;/li&gt;
&lt;li&gt;Problem-solving approaches&lt;/li&gt;
&lt;li&gt;Availability and collaboration preferences&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;2. Intelligent portfolios&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;The agent organizes collected information into structured portfolios, including a short summary of each user's approach to problem-solving.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. Opportunity discovery&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;STRANGER recommends five to seven personalized directions, such as:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Hackathon teammates&lt;/li&gt;
&lt;li&gt;Research collaborations&lt;/li&gt;
&lt;li&gt;Internships and employment&lt;/li&gt;
&lt;li&gt;Startup partnerships&lt;/li&gt;
&lt;li&gt;Technical mentorship&lt;/li&gt;
&lt;li&gt;Open-source collaborations&lt;/li&gt;
&lt;li&gt;Skill-exchange opportunities&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;4. Compatibility matching&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;The matching engine is designed to compare profiles according to technical requirements, relevant experience, goals, availability, and complementary skills.&lt;/p&gt;

&lt;p&gt;It uses a transparent 0–100 scoring rubric rather than treating AI-generated percentages as objective probabilities.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;5. Shared profile network&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;STRANGER's architecture separates fictional demonstration profiles from real user profiles.&lt;/p&gt;

&lt;p&gt;The demo dataset is intended to cover four groups: students, recruiters, companies, and researchers.&lt;/p&gt;

&lt;p&gt;Real-user discovery requires explicit consent. Synthetic profiles must always be identified as demonstrations.&lt;/p&gt;

&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;p&gt;STRANGER is hosted on Fetch.ai Agentverse and is discoverable through ASI:One.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Agent name:&lt;/strong&gt; STRANGER&lt;br&gt;&lt;br&gt;
&lt;strong&gt;Agent handle:&lt;/strong&gt; &lt;code&gt;@stranger-2&lt;/code&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Try it:&lt;/strong&gt;&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Visit &lt;a href="https://asi1.ai" rel="noopener noreferrer"&gt;ASI:One&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;Search for &lt;code&gt;@stranger-2&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;Start a conversation with STRANGER.&lt;/li&gt;
&lt;li&gt;Describe your skills, background, and goals.&lt;/li&gt;
&lt;li&gt;Explore the networking and opportunity recommendations.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;The live prototype has successfully demonstrated ASI:One communication and personalized portfolio responses.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Agentverse demo link:&lt;/strong&gt; &lt;a href="https://asi1.ai/ai/agent1qwxut7yqpcelu26flf6wasw7ckztc2n5vnmgef2xqljktqew24fykyrs9mq" rel="noopener noreferrer"&gt;https://asi1.ai/ai/agent1qwxut7yqpcelu26flf6wasw7ckztc2n5vnmgef2xqljktqew24fykyrs9mq&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Video demonstration:&lt;/strong&gt; &lt;/p&gt;

&lt;p&gt;The persistent demo database, fully automated matching, and connection-request workflows are still being validated. I am keeping these distinctions clear rather than presenting planned functionality as completed.&lt;/p&gt;
&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;STRANGER is developed in Python using Fetch.ai's open-source uAgents framework.&lt;/p&gt;

&lt;p&gt;Its main components include:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;agent.py&lt;/code&gt; — agent initialization and protocol registration&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;protocols.py&lt;/code&gt; — chat handling, onboarding, state management, and commands&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;http_client.py&lt;/code&gt; — ASI-1 model integration&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;matching.py&lt;/code&gt; — synthetic profile creation and deterministic compatibility calculations&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The agent implements the Agent Chat Protocol and publishes its protocol manifest for ASI:One compatibility.&lt;/p&gt;

&lt;p&gt;Example:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="kn"&gt;from&lt;/span&gt; &lt;span class="n"&gt;uagents&lt;/span&gt; &lt;span class="kn"&gt;import&lt;/span&gt; &lt;span class="n"&gt;Agent&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;Context&lt;/span&gt;
&lt;span class="kn"&gt;from&lt;/span&gt; &lt;span class="n"&gt;protocols&lt;/span&gt; &lt;span class="kn"&gt;import&lt;/span&gt; &lt;span class="n"&gt;chat_protocol&lt;/span&gt;

&lt;span class="n"&gt;agent&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nc"&gt;Agent&lt;/span&gt;&lt;span class="p"&gt;()&lt;/span&gt;

&lt;span class="nd"&gt;@agent.on_event&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;startup&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="k"&gt;async&lt;/span&gt; &lt;span class="k"&gt;def&lt;/span&gt; &lt;span class="nf"&gt;on_startup&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;ctx&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="n"&gt;Context&lt;/span&gt;&lt;span class="p"&gt;):&lt;/span&gt;
    &lt;span class="n"&gt;ctx&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;logger&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;info&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;
        &lt;span class="sa"&gt;f&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;STRANGER is online at &lt;/span&gt;&lt;span class="si"&gt;{&lt;/span&gt;&lt;span class="n"&gt;ctx&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;agent&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;address&lt;/span&gt;&lt;span class="si"&gt;}&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;
    &lt;span class="p"&gt;)&lt;/span&gt;

&lt;span class="n"&gt;agent&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;include&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;chat_protocol&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;publish_manifest&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="bp"&gt;True&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;

&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;__name__&lt;/span&gt; &lt;span class="o"&gt;==&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;__main__&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="n"&gt;agent&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;run&lt;/span&gt;&lt;span class="p"&gt;()&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;This allows STRANGER to operate as a discoverable Fetch.ai agent.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Open-source framework:&lt;/strong&gt; &lt;a href="https://asi1.ai/ai/agent1qwxut7yqpcelu26flf6wasw7ckztc2n5vnmgef2xqljktqew24fykyrs9mq" rel="noopener noreferrer"&gt;https://asi1.ai/ai/agent1qwxut7yqpcelu26flf6wasw7ckztc2n5vnmgef2xqljktqew24fykyrs9mq&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;I developed STRANGER within the Fetch.ai ecosystem, using Agentverse for agent hosting and ASI:One for conversational interaction.&lt;/p&gt;

&lt;h3&gt;
  
  
  Technology stack
&lt;/h3&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Technology&lt;/th&gt;
&lt;th&gt;Purpose&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Python&lt;/td&gt;
&lt;td&gt;Agent implementation and matching logic&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;uAgents&lt;/td&gt;
&lt;td&gt;Open-source agent framework&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Agent Chat Protocol&lt;/td&gt;
&lt;td&gt;Structured agent communication&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Agentverse&lt;/td&gt;
&lt;td&gt;Hosting, registration, and persistent storage&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;ASI:One&lt;/td&gt;
&lt;td&gt;Conversational user interface&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;ASI-1&lt;/td&gt;
&lt;td&gt;Language understanding and portfolio generation&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Deterministic Python&lt;/td&gt;
&lt;td&gt;Workflow control and compatibility scoring&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h3&gt;
  
  
  Architecture
&lt;/h3&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;User
  ↓
ASI:One
  ↓
STRANGER uAgent
  ↓
Adaptive Interview
  ↓
Portfolio Intelligence
  ↓
Opportunity Recommendations
  ↓
Profile Registry
  ↓
Compatibility Engine
  ↓
Ranked Connections
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  The biggest technical challenge
&lt;/h3&gt;

&lt;p&gt;My first implementation behaved too much like a conventional chatbot.&lt;/p&gt;

&lt;p&gt;It could understand a user's profile and offer useful advice, but it sometimes continued asking questions even after enough information was available.&lt;/p&gt;

&lt;p&gt;That revealed an important engineering lesson:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;An intelligent response is not the same as an intelligent workflow.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;To address this, I began separating AI reasoning from deterministic execution.&lt;/p&gt;

&lt;p&gt;Python manages explicit workflow states, question limits, storage operations, and compatibility calculations.&lt;/p&gt;

&lt;p&gt;ASI-1 handles tasks that benefit from language understanding and generation.&lt;/p&gt;

&lt;p&gt;This separation also helps reduce model-token consumption on the Agentverse Basic Plan.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;Open innovation matters because autonomous agents should be understandable, extensible, and interoperable.&lt;/p&gt;

&lt;p&gt;The open-source uAgents framework allows developers to inspect communication patterns, implement custom protocols, and build agents without depending entirely on a proprietary visual builder.&lt;/p&gt;

&lt;p&gt;For STRANGER, that means I can customize:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;How agents communicate&lt;/li&gt;
&lt;li&gt;How interviews progress&lt;/li&gt;
&lt;li&gt;How user information is structured&lt;/li&gt;
&lt;li&gt;How matching scores are calculated&lt;/li&gt;
&lt;li&gt;How profiles are stored and retrieved&lt;/li&gt;
&lt;li&gt;How external capabilities can be integrated&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The project uses ASI-1 as a hosted reasoning model, so it is not a fully open-weight AI application.&lt;/p&gt;

&lt;p&gt;However, its agent framework, deterministic matching logic, and workflow implementation demonstrate how open-source tooling can support an extensible agent-based system.&lt;/p&gt;

&lt;p&gt;The longer-term goal is a network of independently useful agents that can collaborate through shared protocols.&lt;/p&gt;

&lt;h2&gt;
  
  
  My Agent Session
&lt;/h2&gt;

&lt;p&gt;My development journey involved generating the initial STRANGER agent, testing it through ASI:One, inspecting its Python source, and iteratively improving its onboarding and matching architecture.&lt;/p&gt;

&lt;p&gt;Several early tests exposed issues involving repeated questions, missing persistent profiles, and model-token limits.&lt;/p&gt;

&lt;p&gt;These failures helped identify where deterministic state management and explicit data verification were necessary.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Agent session:&lt;/strong&gt; [Add DevRelay session URL or embed, if available]&lt;/p&gt;

&lt;h2&gt;
  
  
  The Spirit of It
&lt;/h2&gt;

&lt;p&gt;The most interesting part of STRANGER isn't simply generating another AI profile.&lt;/p&gt;

&lt;p&gt;It's the possibility of discovering someone who could change the direction of a project, career, or idea.&lt;/p&gt;

&lt;p&gt;Maybe a student in Bengaluru has the technical skills a researcher has been searching for.&lt;/p&gt;

&lt;p&gt;Maybe a founder has an idea but needs a developer willing to build it.&lt;/p&gt;

&lt;p&gt;Maybe two strangers have complementary strengths but would never have discovered each other.&lt;/p&gt;

&lt;p&gt;That's the ghost I wanted to bring into the machine: the invisible potential of human connections.&lt;/p&gt;

&lt;p&gt;STRANGER is still evolving, but its purpose remains the same.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Everyone is someone's missing connection.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;And maybe the right connection is already waiting somewhere in the network.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;p&gt;Entering the overall Hacktoberfest Open-Source AI Challenge.&lt;/p&gt;

&lt;p&gt;I am not claiming a partner-specific prize category unless the corresponding technology is meaningfully integrated and demonstrated.&lt;/p&gt;




&lt;p&gt;&lt;strong&gt;Built by:&lt;/strong&gt; K M Lochan&lt;br&gt;&lt;br&gt;
&lt;strong&gt;University:&lt;/strong&gt; Dayananda Sagar University, Bengaluru&lt;br&gt;&lt;br&gt;
&lt;strong&gt;Project:&lt;/strong&gt; STRANGER&lt;br&gt;&lt;br&gt;
&lt;strong&gt;Agentverse handle:&lt;/strong&gt; &lt;code&gt;@stranger-2&lt;/code&gt;&lt;br&gt;&lt;br&gt;
&lt;strong&gt;Built with:&lt;/strong&gt; Python, uAgents, Agentverse, ASI:One, ASI-1&lt;/p&gt;

&lt;h1&gt;
  
  
  Hacktoberfest #OpenSource #AI #Python
&lt;/h1&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
      <category>hacktoberfest</category>
      <category>ai</category>
    </item>
    <item>
      <title>Margin: Lazy Way to Solve Math with Open-Source AI</title>
      <dc:creator>lochan k m</dc:creator>
      <pubDate>Sun, 25 Oct 2026 13:56:05 +0000</pubDate>
      <link>https://dev.to/lochan_km_8f2c3f2051e9d1/margin-lazy-way-to-solve-math-with-open-source-ai-5e2n</link>
      <guid>https://dev.to/lochan_km_8f2c3f2051e9d1/margin-lazy-way-to-solve-math-with-open-source-ai-5e2n</guid>
      <description>&lt;p&gt;&lt;strong&gt;&lt;u&gt;What I Built&lt;/u&gt;&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Doing math on a computer has always felt like unnecessary hard work. If you want to solve an equation, you usually have to fight with equation editors, type out awkward LaTeX syntax, or jump across three different tabs between a graphing calculator, a textbook, and a scratchpad. Even with pen and paper, the moment you make a small mistake on line two of a long derivation, you end up wasting twenty minutes doing tedious algebra before realizing your answer is completely off.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;&lt;em&gt;I built Margin because I wanted to be lazy about math.&lt;/em&gt;&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Code&lt;/p&gt;

&lt;p&gt;The complete source code for Margin, including the canvas engine, the SymPy verification service, and the vision pipeline, is available on &lt;strong&gt;GitHub:&lt;/strong&gt; &lt;a href="https://github.com/lochanharishwar/MYPREP" rel="noopener noreferrer"&gt;https://github.com/lochanharishwar/MYPREP&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Margin is an infinite digital whiteboard where you do not have to type anything. You just pick up a stylus or use your mouse, scribble down a problem naturally, and let open-source AI and symbolic computing do the rest. As soon as you pause, Margin reads your handwriting, verifies every algebraic step with a deterministic math engine, plots interactive coordinate graphs, and speaks back to you with a patient, friendly conversational tutor that explains where you went wrong. You do not even have to read a giant wall of text because the tutor talks through the lesson out loud while writing the verified steps right onto the board for you.&lt;/p&gt;

&lt;p&gt;&lt;u&gt;Demo&lt;/u&gt;&lt;/p&gt;

&lt;p&gt;You can try out the live web application here: &lt;a href="https://gen-lang-client-0249652862.web.app/notebook/default" rel="noopener noreferrer"&gt;https://gen-lang-client-0249652862.web.app/notebook/default&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;You can also run the entire application locally on your machine with the Next.js frontend and the local Python FastAPI microservice.&lt;/p&gt;

&lt;p&gt;How I Built It&lt;/p&gt;

&lt;p&gt;Margin is built around a clean separation between generative open-weight vision and deterministic symbolic math:&lt;/p&gt;

&lt;p&gt;1.&lt;strong&gt;The Vector Canvas:&lt;/strong&gt; Built with Next.js 14, React, and an HTML5 Canvas that captures handwritten coordinates and smooths raw points into anti-aliased quadratic curves in real time.&lt;/p&gt;

&lt;p&gt;2.&lt;strong&gt;Open-Weight Vision and Handwriting OCR:&lt;/strong&gt; When you pause your writing, the canvas isolates the bounding box of your strokes and converts it into a lightweight, high-contrast image. &lt;/p&gt;

&lt;p&gt;This gets processed by Google Gemma 4 and fast multimodal models configured specifically for character disambiguation, distinguishing things like the variable x from multiplication, or the letter t from a plus sign.&lt;/p&gt;

&lt;p&gt;3.&lt;strong&gt;Deterministic Verification with SymPy:&lt;/strong&gt; Generative models are great at reading pictures, but they are notoriously unreliable at arithmetic. &lt;/p&gt;

&lt;p&gt;To keep the math strictly correct, Margin sends every recognized equation to a local Python FastAPI service running SymPy. &lt;/p&gt;

&lt;p&gt;SymPy evaluates algebraic equivalence, computes exact derivatives, finds roots, and confirms the derivation step by step.&lt;/p&gt;

&lt;p&gt;4.&lt;strong&gt;Interactive Display and Graphs&lt;/strong&gt;: The frontend uses KaTeX to render clean math cards and MathJS to render interactive Cartesian coordinate graphs beside your handwritten work.&lt;/p&gt;

&lt;p&gt;5.&lt;strong&gt;Conversational Voice Tutor:&lt;/strong&gt; Mathematical explanations are phonetized into natural conversational English and vocalized using streaming neural speech synthesis, giving you a real-time teaching assistant that speaks like a patient instructor rather than a robotic text reader.&lt;br&gt;
Why Does Open Innovation Matter?&lt;/p&gt;

&lt;p&gt;6.&lt;strong&gt;Open innovation was not just an option for this project, it was the only way to make it work reliably.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;First, closed commercial LLMs try to do everything inside a single black box. When you ask a closed model to solve a multi-step calculus problem, it often hallucinates plausible-looking numbers that are mathematically wrong. Open-source innovation allows us to combine the best open-weight vision models with SymPy, a mature, deterministic computer algebra system that has been refined by the global scientific community for over a decade. That pairing gives you actual mathematical proof instead of a statistical guess.&lt;/p&gt;

&lt;p&gt;Second, open innovation protects privacy and keeps costs at zero. When students and engineers work on personal scratchpads, their handwritten notes and ideas should not be fed into a proprietary training pipeline or locked behind a twenty-dollar monthly subscription. Because Margin is built on open standards, lightweight SQLite, and modular Python services, anyone can clone the repo, run it on their own laptop, and have a permanent personal math tutor without relying on closed corporate servers.&lt;/p&gt;

&lt;p&gt;7./&lt;strong&gt;My Agent Session&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;This project was built and debugged collaboratively using an autonomous coding agent session. The agent assisted in scaffolding the Next.js 14 App Router, writing the vector stroke rasterizer, fine-tuning the SymPy verification microservice, and solving audio race conditions with AbortControllers so speech never overlaps when drawing quickly.&lt;/p&gt;

&lt;p&gt;8.&lt;strong&gt;The Lazy Payoff&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;The lazy payoff is huge. Instead of spending ten minutes formatting an integral in a document editor or getting stuck on line four of a quadratic formula because of a missed negative sign, you just &lt;u&gt;&lt;strong&gt;write the problem down and sit back. Margin instantly tells you if your intermediate step is algebraically equivalent to your previous one, graphs the curve automatically so you do not have to plot points by hand, and reads the answer out loud.&lt;/strong&gt;&lt;/u&gt; It completely eliminates the tedious busywork of math, letting you move slow, stay relaxed, and focus purely on understanding the concept.&lt;/p&gt;

&lt;p&gt;Prize Categories&lt;br&gt;
-Overall Open-Source AI Challenge&lt;br&gt;
-Best Local or Offline-First AI Application&lt;br&gt;
-Best Educational or Productivity Tool&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
      <category>ai</category>
    </item>
    <item>
      <title>Margin: The Lazy Way to Solve Math with Open-Source AI</title>
      <dc:creator>lochan k m</dc:creator>
      <pubDate>Sun, 25 Oct 2026 13:51:02 +0000</pubDate>
      <link>https://dev.to/lochan_km_8f2c3f2051e9d1/margin-the-lazy-way-to-solve-math-with-open-source-ai-1jaj</link>
      <guid>https://dev.to/lochan_km_8f2c3f2051e9d1/margin-the-lazy-way-to-solve-math-with-open-source-ai-1jaj</guid>
      <description>&lt;p&gt;&lt;strong&gt;&lt;u&gt;What I Built&lt;/u&gt;&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Doing math on a computer has always felt like unnecessary hard work. If you want to solve an equation, you usually have to fight with equation editors, type out awkward LaTeX syntax, or jump across three different tabs between a graphing calculator, a textbook, and a scratchpad. Even with pen and paper, the moment you make a small mistake on line two of a long derivation, you end up wasting twenty minutes doing tedious algebra before realizing your answer is completely off.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;&lt;em&gt;I built Margin because I wanted to be lazy about math.&lt;/em&gt;&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Code&lt;/p&gt;

&lt;p&gt;The complete source code for Margin, including the canvas engine, the SymPy verification service, and the vision pipeline, is available on &lt;strong&gt;GitHub:&lt;/strong&gt; &lt;a href="https://github.com/lochanharishwar/MYPREP" rel="noopener noreferrer"&gt;https://github.com/lochanharishwar/MYPREP&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Margin is an infinite digital whiteboard where you do not have to type anything. You just pick up a stylus or use your mouse, scribble down a problem naturally, and let open-source AI and symbolic computing do the rest. As soon as you pause, Margin reads your handwriting, verifies every algebraic step with a deterministic math engine, plots interactive coordinate graphs, and speaks back to you with a patient, friendly conversational tutor that explains where you went wrong. You do not even have to read a giant wall of text because the tutor talks through the lesson out loud while writing the verified steps right onto the board for you.&lt;/p&gt;

&lt;p&gt;&lt;u&gt;Demo&lt;/u&gt;&lt;/p&gt;

&lt;p&gt;You can try out the live web application here: &lt;a href="https://gen-lang-client-0249652862.web.app/notebook/default" rel="noopener noreferrer"&gt;https://gen-lang-client-0249652862.web.app/notebook/default&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;You can also run the entire application locally on your machine with the Next.js frontend and the local Python FastAPI microservice.&lt;/p&gt;

&lt;p&gt;How I Built It&lt;/p&gt;

&lt;p&gt;Margin is built around a clean separation between generative open-weight vision and deterministic symbolic math:&lt;/p&gt;

&lt;p&gt;1.&lt;strong&gt;The Vector Canvas:&lt;/strong&gt; Built with Next.js 14, React, and an HTML5 Canvas that captures handwritten coordinates and smooths raw points into anti-aliased quadratic curves in real time.&lt;/p&gt;

&lt;p&gt;2.&lt;strong&gt;Open-Weight Vision and Handwriting OCR:&lt;/strong&gt; When you pause your writing, the canvas isolates the bounding box of your strokes and converts it into a lightweight, high-contrast image. &lt;/p&gt;

&lt;p&gt;This gets processed by Google Gemma 4 and fast multimodal models configured specifically for character disambiguation, distinguishing things like the variable x from multiplication, or the letter t from a plus sign.&lt;/p&gt;

&lt;p&gt;3.&lt;strong&gt;Deterministic Verification with SymPy:&lt;/strong&gt; Generative models are great at reading pictures, but they are notoriously unreliable at arithmetic. &lt;/p&gt;

&lt;p&gt;To keep the math strictly correct, Margin sends every recognized equation to a local Python FastAPI service running SymPy. &lt;/p&gt;

&lt;p&gt;SymPy evaluates algebraic equivalence, computes exact derivatives, finds roots, and confirms the derivation step by step.&lt;/p&gt;

&lt;p&gt;4.&lt;strong&gt;Interactive Display and Graphs&lt;/strong&gt;: The frontend uses KaTeX to render clean math cards and MathJS to render interactive Cartesian coordinate graphs beside your handwritten work.&lt;/p&gt;

&lt;p&gt;5.&lt;strong&gt;Conversational Voice Tutor:&lt;/strong&gt; Mathematical explanations are phonetized into natural conversational English and vocalized using streaming neural speech synthesis, giving you a real-time teaching assistant that speaks like a patient instructor rather than a robotic text reader.&lt;br&gt;
Why Does Open Innovation Matter?&lt;/p&gt;

&lt;p&gt;6.&lt;strong&gt;Open innovation was not just an option for this project, it was the only way to make it work reliably.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;First, closed commercial LLMs try to do everything inside a single black box. When you ask a closed model to solve a multi-step calculus problem, it often hallucinates plausible-looking numbers that are mathematically wrong. Open-source innovation allows us to combine the best open-weight vision models with SymPy, a mature, deterministic computer algebra system that has been refined by the global scientific community for over a decade. That pairing gives you actual mathematical proof instead of a statistical guess.&lt;/p&gt;

&lt;p&gt;Second, open innovation protects privacy and keeps costs at zero. When students and engineers work on personal scratchpads, their handwritten notes and ideas should not be fed into a proprietary training pipeline or locked behind a twenty-dollar monthly subscription. Because Margin is built on open standards, lightweight SQLite, and modular Python services, anyone can clone the repo, run it on their own laptop, and have a permanent personal math tutor without relying on closed corporate servers.&lt;/p&gt;

&lt;p&gt;7./&lt;strong&gt;My Agent Session&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;This project was built and debugged collaboratively using an autonomous coding agent session. The agent assisted in scaffolding the Next.js 14 App Router, writing the vector stroke rasterizer, fine-tuning the SymPy verification microservice, and solving audio race conditions with AbortControllers so speech never overlaps when drawing quickly.&lt;/p&gt;

&lt;p&gt;8.&lt;strong&gt;The Lazy Payoff&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;The lazy payoff is huge. Instead of spending ten minutes formatting an integral in a document editor or getting stuck on line four of a quadratic formula because of a missed negative sign, you just &lt;u&gt;&lt;strong&gt;write the problem down and sit back. Margin instantly tells you if your intermediate step is algebraically equivalent to your previous one, graphs the curve automatically so you do not have to plot points by hand, and reads the answer out loud.&lt;/strong&gt;&lt;/u&gt; It completely eliminates the tedious busywork of math, letting you move slow, stay relaxed, and focus purely on understanding the concept.&lt;/p&gt;

&lt;p&gt;Prize Categories&lt;br&gt;
-Overall Open-Source AI Challenge&lt;br&gt;
-Best Local or Offline-First AI Application&lt;br&gt;
-Best Educational or Productivity Tool&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
  </channel>
</rss>
