Back to work

Featured · Web/AI

FlashFind

Full-Stack / ML Developer

Hyperlocal AI marketplace: users submit natural-language "Flash Requests" and the system parses intent with an LLM then Smart-Pings nearby students ranked by a Random Forest to fulfill needs in minutes.

Overview

Inspiration

Built to solve urgent student needs (forgotten charger, last-minute textbook) by turning campus inventory into an on-demand network — faster than fragmented marketplaces and better for community reuse.

What it does

Students submit a short natural-language Flash Request; an LLM parses intent and item, and a lightweight ML classifier ranks and Smart-Pings nearby students most likely to help. The flow is optimized for speed, low noise, and privacy.

How we built it

Two-step AI pipeline: Google Gemini for language-to-JSON parsing, then a scikit-learn Random Forest (saved as a .joblib) for fast logical matching. Backend is FastAPI (Python) with MongoDB, frontend is React + TypeScript.

Challenges & wins

Key challenges were data alignment with training vs live data and solving cold-start via a synthetic-data factory. We're proud of integrating the Two-Step AI and training a custom Random Forest from synthetic examples.

What's next

Scale to other communities and emergency scenarios where rapid resource distribution matters.

Stack
PythonFastAPIMongoDBGeminiReactTailwind