Hacktoberfest Weekend Challenge: Build for a Friend

palakshi verma — Mon, 05 Oct 2026 06:22:46 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

I built DevDraft, an AI-powered communication agent designed for a friend who recently started their first software engineering internship.

Like many junior developers, my friend suffers from intense imposter syndrome when communicating in professional engineering channels. They were spending 30–45 minutes drafting a single daily standup update or a GitHub Pull Request description. The engineering work wasn't the bottleneck—it was the anxiety of getting the tone, structure, and technical jargon right in front of senior engineers.

DevDraft acts as a patient "Senior Developer" partner. My friend can dump their raw, chaotic, unfiltered notes into the app (e.g., "i fixed the login bug but the css is still weird so I didn't push that part"). The agent instantly transforms this into a crisp, confident, and professional Standup update or a perfectly structured PR template, stripping away apologetic language and formatting it for Slack or GitHub.

Demo

Live Demo: https://devdraft.onrender.com/

(Note: The live demo uses the Groq API for speed, but the app is built to be run locally!)

Code

https://github.com/PalakshiVerma/DevDraft

How I Built It

I decoupled the architecture into a Streamlit frontend and a FastAPI backend to keep the LLM logic flexible.

At the core of the project is the Meta Llama 3.1 (8B) open-weight model. The FastAPI backend receives the raw text and injects it into highly specific system prompts designed to mimic a confident senior engineer.

I deployed the FastAPI backend as a Web Service on Render. Render's automated deployment from GitHub made it incredibly easy to ship the API in minutes, providing a stable endpoint for the Streamlit UI to interact with.

Why Does Open Innovation Matter?

For an intern, using closed AI APIs is a massive security risk. Pasting proprietary company code or internal architecture notes into a closed cloud AI violates almost every company NDA.

This is where open innovation completely outshines closed systems. I built DevDraft with a "Strictly Local" mode. Because it utilizes open-weight models, my friend can run the exact same application entirely offline using Ollama.

By running Llama 3.1 locally, my friend gets senior-level code review and communication polish without a single line of their company's proprietary code ever leaving their laptop. A closed API would make this tool unusable for its intended audience; open-source AI makes it secure, free, and completely private.

Prize Categories

Best Use of Render: The core AI inference API (built with FastAPI) is hosted and deployed using Render's Web Services, acting as the bridge between the UI and the open-weight LLM.