<?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: Adarsh Negi</title>
    <description>The latest articles on DEV Community by Adarsh Negi (@adarshnegi).</description>
    <link>https://dev.to/adarshnegi</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%2F4159268%2Fc1a302ea-840a-4c28-a6e6-d4d1c98548f1.jpg</url>
      <title>DEV Community: Adarsh Negi</title>
      <link>https://dev.to/adarshnegi</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/adarshnegi"/>
    <language>en</language>
    <item>
      <title>TrailMate 🌿👻 — A Ghost in Your Pocket That Wants You Outside</title>
      <dc:creator>Adarsh Negi</dc:creator>
      <pubDate>Tue, 27 Oct 2026 10:14:57 +0000</pubDate>
      <link>https://dev.to/adarshnegi/trailmate-a-ghost-in-your-pocket-that-wants-you-outside-30le</link>
      <guid>https://dev.to/adarshnegi/trailmate-a-ghost-in-your-pocket-that-wants-you-outside-30le</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;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;h3&gt;
  
  
  TrailMate 🌿👻 — A Ghost in Your Pocket That Wants You Outside
&lt;/h3&gt;

&lt;p&gt;What if the ghost in your machine didn't want to haunt you, steal your data, or keep you doomscrolling?&lt;/p&gt;

&lt;p&gt;What if it wanted you to close your laptop and go outside?&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;TrailMate is a local-AI-powered outdoor exploration companion that turns ordinary walks into small adventures.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Choose an environment, set a duration and difficulty, and TrailMate generates an outdoor mission with three objectives. These missions encourage you to look more closely at your surroundings, discover visual details, and pause for mindful observation.&lt;/p&gt;

&lt;p&gt;A neighborhood you've walked through a hundred times might still have something you've never noticed. A quiet park might tell a different story when you slow down and pay attention.&lt;/p&gt;

&lt;p&gt;That's the spirit behind this project: treating the familiar world as something waiting to be rediscovered.&lt;/p&gt;

&lt;p&gt;For the &lt;em&gt;Ghost in the Machine&lt;/em&gt; theme, I like the idea of an invisible companion that quietly nudges you toward the physical world rather than demanding more of your attention.&lt;/p&gt;

&lt;p&gt;The ghost stays in the machine. The adventure happens outside it.&lt;/p&gt;

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

&lt;p&gt;🎥 &lt;strong&gt;Video demo:&lt;/strong&gt; [Add your demo link here]&lt;/p&gt;

&lt;p&gt;TrailMate currently runs locally with a Next.js frontend, a FastAPI backend, and Gemma 3 4B through Ollama.&lt;/p&gt;

&lt;p&gt;The demo will show the mission preferences, local AI generation, and the resulting three-objective mission.&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;💻 &lt;strong&gt;GitHub repository:&lt;/strong&gt; [Add your public GitHub repository URL here]&lt;/p&gt;

&lt;p&gt;The project is built with:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Next.js, React, TypeScript, and Tailwind CSS&lt;/strong&gt; for the frontend.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Python and FastAPI&lt;/strong&gt; for the backend.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Gemma 3 4B through Ollama&lt;/strong&gt; for local AI inference.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Pydantic&lt;/strong&gt; for validating structured model output.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Pytest and Node's test runner&lt;/strong&gt; for automated tests.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The architecture keeps the AI integration behind the backend rather than exposing the model runtime directly to the frontend.&lt;/p&gt;

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

&lt;p&gt;I started by building the backend, integrating Ollama, and defining a structured mission schema. Then I connected the frontend to the mission-generation API.&lt;/p&gt;

&lt;p&gt;The current flow is straightforward:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;The user selects an environment, difficulty, and duration.&lt;/li&gt;
&lt;li&gt;The frontend sends those preferences to FastAPI.&lt;/li&gt;
&lt;li&gt;The backend asks Gemma 3 4B to generate a structured mission.&lt;/li&gt;
&lt;li&gt;The backend validates the response before returning it.&lt;/li&gt;
&lt;li&gt;The frontend displays the mission and its three objectives.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;One of the most interesting challenges has been running a 4-billion-parameter model on my own laptop's CPU.&lt;/p&gt;

&lt;p&gt;Local inference is not instant. Cold starts and generation time can make a simple interaction feel surprisingly long, which means timeout configuration, loading states, and failure recovery matter just as much as the model call itself.&lt;/p&gt;

&lt;p&gt;That experience has reminded me that building an AI product isn't simply about getting a model to produce an answer. It's about making the whole system useful and reliable despite the limitations of the hardware.&lt;/p&gt;

&lt;p&gt;TrailMate is still evolving. The current focus is making mission generation work reliably; features such as camera-based objective verification and a persistent discovery journal remain future work.&lt;/p&gt;

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

&lt;p&gt;For TrailMate, open innovation isn't just a way to save API costs. It's an opportunity to build an AI experience that I can run and understand on my own hardware.&lt;/p&gt;

&lt;p&gt;Using Gemma 3 4B through Ollama lets me experiment with local inference instead of making a closed, cloud-hosted AI API a mandatory dependency.&lt;/p&gt;

&lt;p&gt;That gives the project several advantages:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Control:&lt;/strong&gt; I can configure the model runtime and decide how the application communicates with it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Privacy potential:&lt;/strong&gt; Mission preferences can be processed locally rather than automatically being sent to a third-party AI service.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Flexibility:&lt;/strong&gt; The model integration can evolve as I experiment with different models and configurations.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Accessibility for experimentation:&lt;/strong&gt; I can learn about prompts, structured generation, validation, and inference without needing to build every experiment around a paid API.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;There are trade-offs, of course. My laptop's CPU is considerably slower than dedicated inference hardware, and local inference still requires the model and runtime to be installed. I wouldn't claim that every part of TrailMate currently works offline without testing that end to end.&lt;/p&gt;

&lt;p&gt;But open models make this kind of experimentation possible. I can inspect the application architecture, control the inference setup, and learn from the constraints instead of treating the model as a black box behind someone else's API.&lt;/p&gt;

&lt;p&gt;That freedom is what makes open innovation valuable to this project.&lt;/p&gt;

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

&lt;p&gt;[Add a DevRelay agent session link here, if available.]&lt;/p&gt;

&lt;p&gt;I'd like the development process to be part of the story too: not just the finished interface, but the iterations, debugging, and engineering decisions that made it work.&lt;/p&gt;

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

&lt;p&gt;Halloween usually gives us ghosts that want our attention.&lt;/p&gt;

&lt;p&gt;TrailMate imagines a different kind of ghost: one that reminds us how much of the world we miss when our attention is permanently trapped inside a screen.&lt;/p&gt;

&lt;p&gt;It might be the old tree on your street, a strange pattern in a wall, the sound of birds somewhere nearby, or a corner of your neighborhood you've never explored.&lt;/p&gt;

&lt;p&gt;You don't need a supernatural encounter to rediscover something. Sometimes you just need a reason to look.&lt;/p&gt;

&lt;p&gt;And perhaps that's a fitting use for AI: not to replace the experience of being present, but to help us find our way back to it.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The best TrailMate session ends with the user putting their phone away.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;The ghost lives in the machine. The story happens out there. 🌿👻&lt;/p&gt;

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

&lt;p&gt;[Add the applicable partner categories after checking the official challenge rules.]&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
      <category>ai</category>
      <category>opensource</category>
    </item>
    <item>
      <title>FreelanceFlow: A Private AI Assistant for Freelancers</title>
      <dc:creator>Adarsh Negi</dc:creator>
      <pubDate>Tue, 20 Oct 2026 12:34:02 +0000</pubDate>
      <link>https://dev.to/adarshnegi/freelanceflow-a-private-ai-assistant-for-freelancers-4d56</link>
      <guid>https://dev.to/adarshnegi/freelanceflow-a-private-ai-assistant-for-freelancers-4d56</guid>
      <description>&lt;h1&gt;
  
  
  FreelanceFlow: A Private AI Assistant for Freelancers
&lt;/h1&gt;

&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week2-2026-10-12"&gt;Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

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

&lt;p&gt;Freelancers shouldn't need to spend their evenings organizing receipts, extracting invoice details, and preparing expense reports.&lt;/p&gt;

&lt;p&gt;That's why I'm building &lt;strong&gt;FreelanceFlow&lt;/strong&gt; — a local-first AI assistant that helps freelancers organize their financial documents without relying on a cloud-based AI API.&lt;/p&gt;

&lt;p&gt;The idea is simple: drop in an invoice or receipt, let AI extract the relevant information, review the results, and export organized records for accounting.&lt;/p&gt;

&lt;p&gt;FreelanceFlow is designed to help independent developers, consultants, and small business owners manage repetitive administrative work so they can spend more time on the work that actually pays.&lt;/p&gt;

&lt;p&gt;The planned core features include:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Smart document extraction:&lt;/strong&gt; Extract vendor names, dates, amounts, taxes, and currencies from invoices and receipts.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Expense categorization:&lt;/strong&gt; Organize transactions into useful expense categories.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Human-in-the-loop verification:&lt;/strong&gt; Review extracted information before it enters your records.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Exportable reports:&lt;/strong&gt; Turn scattered documents into structured CSV data for bookkeeping.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Local-first processing:&lt;/strong&gt; Keep sensitive financial documents on the user's own machine.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The business model is straightforward: a small, affordable desktop application for freelancers who want to save time without paying for another recurring AI subscription.&lt;/p&gt;

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

&lt;p&gt;[Add your deployed demo or a short screen-recording link here.]&lt;/p&gt;

&lt;p&gt;The demo will show the workflow from importing a document to reviewing extracted fields and exporting structured records.&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://github.com/" rel="noopener noreferrer"&gt;GitHub Repository — FreelanceFlow&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Replace this placeholder with the actual repository URL before publishing.&lt;/em&gt;&lt;/p&gt;

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

&lt;p&gt;The central design decision is to use open-weight AI as a core part of the application rather than treating it as an optional feature.&lt;/p&gt;

&lt;p&gt;My planned stack includes:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Ollama:&lt;/strong&gt; Runs the language model locally.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Gemma 3 4B:&lt;/strong&gt; Provides local inference for document understanding and expense categorization.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Python and FastAPI:&lt;/strong&gt; Handle document processing, validation, and application logic.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;PyMuPDF and OCR tools:&lt;/strong&gt; Extract text from PDFs and scanned receipts.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;SQLite:&lt;/strong&gt; Stores structured expense records locally.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;React and TypeScript:&lt;/strong&gt; Power the user interface.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The architecture separates probabilistic AI tasks from deterministic financial operations. The model assists with interpreting documents and assigning categories, while Python validates fields and performs calculations.&lt;/p&gt;

&lt;p&gt;That distinction matters. A plausible-looking AI response is not necessarily an accurate financial record, so extracted values need validation and a review step before being accepted.&lt;/p&gt;

&lt;p&gt;The goal is to keep the workflow useful, transparent, and accessible on consumer hardware.&lt;/p&gt;

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

&lt;p&gt;For FreelanceFlow, open innovation isn't just a technology preference. It directly influences whether the product makes sense as a small business tool.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;1. Privacy by design&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Invoices can contain client names, addresses, payment details, and commercially sensitive information. Running inference locally makes it possible to process these documents without sending their contents to an external AI inference provider.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;2. More predictable operating costs&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;A freelancer processing hundreds of documents shouldn't necessarily need to pay an AI API provider for every extraction. Local inference avoids per-request model API charges, although hardware, development, and distribution still have costs.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. Freedom to experiment&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Open-weight models give developers more control over model selection, prompting, evaluation, and deployment. If a different model performs better on scanned invoices or unfamiliar currencies, the application can be adapted without being tied to one proprietary API.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;4. A product that can work without a permanent cloud dependency&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Once the model and required components are installed, the core workflow can be designed to operate offline. That can make the tool more useful for people who want their financial records to remain under their control.&lt;/p&gt;

&lt;p&gt;There are trade-offs. Local models require suitable hardware, and extraction accuracy must be tested across different document formats. A smaller model may also need more careful prompting and validation than a larger hosted model.&lt;/p&gt;

&lt;p&gt;But for a focused task like invoice organization, the ability to control data flow, inference costs, and model choice is a meaningful advantage.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;My takeaway:&lt;/strong&gt; Open-source AI makes it possible to build a small, specialized product without automatically inheriting the recurring costs and data-sharing assumptions of a hosted AI service.&lt;/p&gt;

&lt;h2&gt;
  
  
  My First User
&lt;/h2&gt;

&lt;p&gt;The next milestone is to put FreelanceFlow in front of a freelancer who regularly handles invoices and receipts.&lt;/p&gt;

&lt;p&gt;I'd like to measure three things:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;How much time does it save compared with manual entry?&lt;/li&gt;
&lt;li&gt;How often does the user need to correct extracted fields?&lt;/li&gt;
&lt;li&gt;Would the user pay for a tool that makes this process faster and keeps their documents local?&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;I haven't included fabricated testimonials or payment claims. The goal is to validate the product with a real user and let their feedback determine what gets built next.&lt;/p&gt;

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

&lt;p&gt;[Add the applicable partner categories based on the challenge's official rules and the technologies actually used.]&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>TrailMate</title>
      <dc:creator>Adarsh Negi</dc:creator>
      <pubDate>Mon, 12 Oct 2026 05:54:31 +0000</pubDate>
      <link>https://dev.to/adarshnegi/trailmate-aa6</link>
      <guid>https://dev.to/adarshnegi/trailmate-aa6</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week1-2026-10-05"&gt;Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h1&gt;
  
  
  TrailMate 🌿 — Let AI Plan Your Next Adventure
&lt;/h1&gt;

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

&lt;p&gt;&lt;strong&gt;TrailMate is a local-AI-powered outdoor exploration companion designed to turn screen time into real-world experiences.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;We spend so much time looking at screens that even our breaks often involve another screen. TrailMate takes a different approach: use AI to help you step outside, explore your surroundings, and pay attention to the world around you.&lt;/p&gt;

&lt;p&gt;Users can choose their preferred environment, difficulty level, and mission duration. TrailMate then generates a structured outdoor exploration mission with three objectives, including visual discovery and mindful observation activities.&lt;/p&gt;

&lt;p&gt;Whether you're walking through a neighborhood, visiting a park, or exploring a nature trail, TrailMate gives you a small reason to look beyond your phone.&lt;/p&gt;

&lt;p&gt;The idea behind it is simple:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Use technology to get people away from technology.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;TrailMate is being built with local AI at its core, so the experience doesn't have to depend on sending every request to a closed, cloud-hosted AI API.&lt;/p&gt;

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

&lt;p&gt;The current development version runs locally, with a FastAPI backend, a Next.js frontend, and Ollama powering local AI inference.&lt;/p&gt;

&lt;p&gt;I'll share a demonstration of the mission-generation flow here.&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;The project uses a modular architecture:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Frontend:&lt;/strong&gt; Next.js, React, TypeScript, and Tailwind CSS.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Backend:&lt;/strong&gt; Python and FastAPI.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Open-source AI:&lt;/strong&gt; Gemma 3 4B through Ollama.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Validation:&lt;/strong&gt; Pydantic schemas to validate generated mission data.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Testing:&lt;/strong&gt; Pytest for backend tests and Node's test runner for frontend API tests.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The frontend communicates with the backend, while the backend handles AI requests and validates the generated output.&lt;/p&gt;

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

&lt;p&gt;I built TrailMate incrementally, starting with the backend and local AI integration before building the mission-generation interface.&lt;/p&gt;

&lt;p&gt;The AI component uses &lt;strong&gt;Gemma 3 4B through Ollama&lt;/strong&gt;, running locally rather than relying on a closed, hosted language-model API.&lt;/p&gt;

&lt;p&gt;When a user submits their preferences, the frontend sends a request to the FastAPI backend. The backend constructs a prompt, requests structured JSON output from the model, and validates the response against Pydantic schemas before returning the mission.&lt;/p&gt;

&lt;p&gt;The current implementation includes:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Customizable mission duration, difficulty, and environment.&lt;/li&gt;
&lt;li&gt;AI-generated missions containing exactly three objectives.&lt;/li&gt;
&lt;li&gt;Structured response validation and error handling.&lt;/li&gt;
&lt;li&gt;A responsive interface with loading and recovery states.&lt;/li&gt;
&lt;li&gt;Automated backend and frontend tests.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Running local inference on my laptop has also introduced an interesting engineering challenge: balancing generation time with a usable experience on CPU-only hardware. That has made timeout handling, error recovery, and realistic performance expectations important parts of the project.&lt;/p&gt;

&lt;p&gt;TrailMate is still under development. Photo-based objective verification, persistent mission progress, and a discovery journal are planned extensions rather than features I'm claiming as complete.&lt;/p&gt;

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

&lt;p&gt;Open innovation is what makes TrailMate possible as an experiment I can build, understand, and improve independently.&lt;/p&gt;

&lt;p&gt;Using an open-weight model with local inference gives me the freedom to explore AI without making a proprietary API a mandatory part of the core experience. It also gives me more control over how inference is run and how data flows through the application.&lt;/p&gt;

&lt;p&gt;There are trade-offs. Local models can be slower on limited hardware, and building a reliable application requires more than simply calling a model. I still need validation, sensible error handling, and a user experience that accounts for those limitations.&lt;/p&gt;

&lt;p&gt;But that's part of the opportunity: open AI makes it possible to experiment with ideas, learn from the entire stack, and build tools that can run closer to the people using them.&lt;/p&gt;

&lt;p&gt;For TrailMate, that matters because the product is about reconnecting with the physical world. Local inference is a step toward making that experience less dependent on cloud infrastructure.&lt;/p&gt;

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

&lt;p&gt;&lt;em&gt;Optional:&lt;/em&gt; I'll add my development session here if I publish one through DevRelay.&lt;/p&gt;

&lt;p&gt;TrailMate is a work in progress, but the first core loop is taking shape: &lt;strong&gt;choose your surroundings, generate a mission, and go explore.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;The goal isn't to make people spend more time using TrailMate. It's to give them a good reason to put their phones away.&lt;/p&gt;

&lt;p&gt;🌿 &lt;strong&gt;Build with AI. Explore the real world. Touch grass.&lt;/strong&gt;&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>SpeakUp</title>
      <dc:creator>Adarsh Negi</dc:creator>
      <pubDate>Mon, 05 Oct 2026 02:51:03 +0000</pubDate>
      <link>https://dev.to/adarshnegi/speakup-6j</link>
      <guid>https://dev.to/adarshnegi/speakup-6j</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-weekend-2026-10-01"&gt;Hacktoberfest Weekend Challenge: Build for a Friend&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h1&gt;
  
  
  What I Built
&lt;/h1&gt;

&lt;p&gt;I built &lt;strong&gt;SpeakUp&lt;/strong&gt;, a local AI-powered mock interview coach for my friend Ayan.&lt;/p&gt;

&lt;p&gt;Ayan is preparing for placements and has a frustrating problem: &lt;strong&gt;he knows the answers, but struggles to communicate them during interviews.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;When we discuss technical topics normally, he can explain them well. But when placed in an interview-like situation, he tends to hesitate, use filler words, lose structure, and struggle to communicate his thoughts clearly.&lt;/p&gt;

&lt;p&gt;So I built SpeakUp specifically for him.&lt;/p&gt;

&lt;p&gt;Instead of simply giving him interview questions and showing the "correct" answer, SpeakUp acts as an interviewer:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Question → Ayan answers → AI evaluates → Follow-up question → Feedback&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;The system evaluates things such as:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Clarity&lt;/li&gt;
&lt;li&gt;Relevance&lt;/li&gt;
&lt;li&gt;Answer structure&lt;/li&gt;
&lt;li&gt;Filler-word usage&lt;/li&gt;
&lt;li&gt;Completeness&lt;/li&gt;
&lt;li&gt;Communication quality&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;It then gives actionable feedback and continues the interview with a follow-up question.&lt;/p&gt;

&lt;p&gt;The goal isn't to teach Ayan what to say.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;It's to help him communicate what he already knows.&lt;/strong&gt;&lt;/p&gt;

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

&lt;p&gt;🌐 &lt;strong&gt;Live Demo:&lt;/strong&gt; [NO DEPLOYED LINK]&lt;/p&gt;

&lt;p&gt;Here's a short example of an interview session:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Interviewer:&lt;/strong&gt; Explain the difference between a process and a thread.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Ayan answers the question.&lt;/p&gt;

&lt;p&gt;SpeakUp then analyzes the response and provides feedback such as:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Clarity        7/10
Relevance      9/10
Structure      5/10
Communication  6/10

Filler words:
"um"           × 4
"actually"     × 2

Suggestions:
→ Start with a concise definition.
→ Compare the two concepts explicitly.
→ Finish with a concrete example.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;It then asks a follow-up question instead of ending the conversation.&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;💻 &lt;strong&gt;GitHub:&lt;/strong&gt; [NO GITHUB REPOSITORY]&lt;/p&gt;

&lt;p&gt;The project is built with Python and is designed to run locally.&lt;/p&gt;

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

&lt;p&gt;The core of SpeakUp is an &lt;strong&gt;open-weight AI model running locally&lt;/strong&gt; rather than a closed AI API.&lt;/p&gt;

&lt;p&gt;The current stack is:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Python&lt;/strong&gt; — application logic&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Streamlit&lt;/strong&gt; — user interface&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ollama&lt;/strong&gt; — local model inference&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Gemma&lt;/strong&gt; — open-weight language model&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Git/GitHub&lt;/strong&gt; — source control and project distribution&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The basic architecture looks like this:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;                ┌───────────────────┐
                │    SpeakUp UI     │
                │    Streamlit      │
                └─────────┬─────────┘
                          │
                          ▼
                ┌───────────────────┐
                │ Interview Engine  │
                │     Python        │
                └─────────┬─────────┘
                          │
                          ▼
                ┌───────────────────┐
                │      Ollama       │
                │  Local Inference  │
                └─────────┬─────────┘
                          │
                          ▼
                ┌───────────────────┐
                │      Gemma        │
                │   Open-weight AI  │
                └───────────────────┘
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The model generates interview questions, evaluates responses, provides feedback, and generates follow-up questions.&lt;/p&gt;

&lt;p&gt;The application is designed around the model rather than using AI as an additional feature.&lt;/p&gt;

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

&lt;p&gt;Interview practice can involve personal information: a student's projects, weaknesses, communication difficulties, career plans, and the answers they give during practice sessions.&lt;/p&gt;

&lt;p&gt;I didn't want Ayan's interview practice to depend entirely on sending that information to a closed API.&lt;/p&gt;

&lt;p&gt;With a local open-weight model, the core interaction can happen directly on the user's machine.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Closed API

Ayan
  ↓
Application
  ↓
External API
  ↓
Closed model
  ↓
Response


SpeakUp

Ayan
  ↓
SpeakUp
  ↓
Local inference
  ↓
Open-weight model
  ↓
Feedback
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Open models also give developers more control.&lt;/p&gt;

&lt;p&gt;I can experiment with different models, change prompts and evaluation strategies, run inference locally, and potentially fine-tune the system for interview practice without redesigning the entire application around a proprietary API.&lt;/p&gt;

&lt;p&gt;For this particular project, &lt;strong&gt;open innovation isn't just about using an open model because the challenge asks for it.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;It directly supports the problem I'm trying to solve:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;private, controllable, locally-run interview practice.&lt;/strong&gt;&lt;/p&gt;

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

&lt;p&gt;I used the session to document the development process and show how the project evolved from the initial idea into the final application.&lt;/p&gt;

&lt;h2&gt;
  
  
  What Ayan Thought
&lt;/h2&gt;

&lt;p&gt;After building the first version, I gave it to Ayan and asked him to actually use it for a mock interview.&lt;/p&gt;

&lt;p&gt;[ADD ACTUAL EXPERIENCE / QUOTE FROM AYAN]&lt;/p&gt;

&lt;p&gt;The most interesting part wasn't whether the AI could answer interview questions.&lt;/p&gt;

&lt;p&gt;It was whether it could help Ayan answer them better.&lt;/p&gt;

&lt;p&gt;That's what I wanted to test.&lt;/p&gt;

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

&lt;p&gt;Building SpeakUp changed how I think about AI projects.&lt;/p&gt;

&lt;p&gt;It would have been easy to build another chatbot that asks questions and generates answers.&lt;/p&gt;

&lt;p&gt;But the real problem wasn't:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;"How can AI answer interview questions?"&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;It was:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;"How can AI help someone communicate knowledge they already have?"&lt;/strong&gt;&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;That distinction shaped the entire project.&lt;/p&gt;

&lt;p&gt;It also showed me one of the practical advantages of open-weight AI: &lt;strong&gt;the model can become part of the application architecture rather than simply being an external API that the application calls.&lt;/strong&gt;&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>weekendchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>About me !</title>
      <dc:creator>Adarsh Negi</dc:creator>
      <pubDate>Sun, 04 Oct 2026 10:10:45 +0000</pubDate>
      <link>https://dev.to/adarshnegi/about-me--kh8</link>
      <guid>https://dev.to/adarshnegi/about-me--kh8</guid>
      <description>&lt;p&gt;Hey everyone! 👋&lt;/p&gt;

&lt;p&gt;I’m Adarsh, a CSE student from India. I’m currently focused on backend development, Python, DSA, and open source.&lt;/p&gt;

&lt;p&gt;I’m starting to explore open source and looking forward to making my first meaningful contributions.&lt;/p&gt;

&lt;p&gt;I joined DEV.to to document my learning journey, connect with other developers, and hopefully learn from the amazing community here.&lt;/p&gt;

&lt;p&gt;Looking forward to building, contributing, and learning along the way!&lt;/p&gt;

</description>
      <category>backend</category>
      <category>beginners</category>
      <category>opensource</category>
      <category>python</category>
    </item>
  </channel>
</rss>
