This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend
What We Built
We built Algo Whisperer, a strict, Socratic AI programming mentor designed to help developers debug C++ and Python code without ever revealing the final solution.
We built this for my friends and peers who practice competitive programming and grind LeetCode daily. When you get stuck on a difficult algorithmic problem, standard LLMs completely ruin the learning process by immediately spitting out the fully optimized code. Algo Whisperer solves this by acting like a real human tutor—it analyzes your logic, finds the flaw (like an early loop termination or a missed base case), and gives you a targeted hint, forcing you to think critically and actually learn the concept.
Demo
Deployed Link : https://algo-whisperer-cid0.onrender.com
Youtube Demo Video : https://www.youtube.com/watch?v=OqStNkSqWqs
For this Hacktoberfest weekend challenge we built Algo Whisperer—a Socratic AI mentor for competitive programmers. Standard AI tools ruin the learning process by just giving you the answer. Let me show you how ours is different.
Here in the UI, We have pasted a classic algorithmic problem on the left, along with my Python attempt. I've got an intentional bug in my logic.
When we ask the chat for help, our React frontend sends this context to a FastAPI backend. We are using LangGraph to enforce strict guardrails before it hits the open-weight Qwen 2.5 Coder model on Hugging Face.
Notice what comes back: instead of rewriting our code and spoiling the solution, the AI analyzes my logic and gives me a targeted hint about my hash map. It forces me to find the bug myself, making it the perfect private tutor for coding interviews.
Code
https://github.com/Kshitij-ambuskar/Algo_Whisperer.git
How I Built It
The architecture relies entirely on open-source frameworks and open-weight models to ensure privacy and accessibility:
AI & Orchestration: I used LangGraph to build a specialized agent workflow that strictly enforces pedagogical guardrails, ensuring the AI never outputs raw code solutions.
Open-Weight Model: The reasoning engine is powered by Qwen/Qwen2.5-Coder-7B-Instruct, accessed via the serverless Hugging Face Inference API. This model is exceptionally good at algorithmic logic.
Backend: A FastAPI (Python) server manages the stateful LangGraph conversation and API routing.
Frontend: A sleek, dark-mode IDE interface built with React, Vite, and Tailwind CSS.
Why Does Open Innovation Matter?
For students and competitive programmers, open innovation makes high-quality learning accessible. By utilizing LangGraph and Hugging Face's open-weight models, we were able to build a tool that costs absolutely zero dollars in API fees to run locally. Furthermore, closed APIs often use your prompts as training data. Open innovation allows developers to debug their code locally and privately, without worrying about exposing proprietary logic or contest solutions to a closed ecosystem.
Prize Categories
- Build for a Friend: Built specifically for my peers practicing data structures and algorithms.
- Open-Source AI / LangGraph: Heavily utilizes LangGraph to constrain the AI's output to hints only.
- Render: The frontend of this AI agent is hosted and deployed using Render's global static site infrastructure.
Top comments (0)