HackaRag

Problem Statement

In hackathon competitions, there's a critical bottleneck: the mentor-to-participant ratio is often too low. Participants waste precious time waiting for mentorship to:✅ Evaluate their project ideas against available bounties ✅ Understand which bounties they should target ✅ Get quantitative feedback on what's working and what needs improvement ✅ Determine if their idea is even viable for the competitionHackaRag solves this by providing an intelligent bridge between mentorship and bounty mapping, using AI to automatically match ideas with relevant bounties and provide comprehensive evaluation.

Solution

We built HackaRag to bridge the gap between hackathon ideas and bounties. The agent scrapes bounty URLs and converts them into JSON. On the app side, we used Python with FAISS for vector search, embedding + chunking pipelines, and an LLM layer for deep analysis. Contestant ideas are matched through multi-level checks against bounties and then evaluated with RAG to generate constructive feedback.The system tells candidates what went well, what went wrong, and what can be improved, while also detecting if the input is a valid idea. We built the UI with Streamlit, containerized everything with Docker, and deployed through Fluence.network for scalability. To keep things efficient, we used caching and staged checks so the LLM only runs on the most relevant matches, making feedback fast and mentor-like even when mentor availability is low.