Left 4 The Reader
LeftForTheReader is an AI-powered mathematics assistant I built to serve as an interactive, context-aware tutor for definitions, theorems, and proofs. The system uses a FastAPI backend connected to a persistent ChromaDB vector database which contains 20k+ ProofWiki entries and a quantized Qwen-2.5-Math-1.5B-Instruct model served via llama.cpp. The app performs retrieval-augmented generation to generate precise, reference-backed mathematical explanations. I containerized the entire stack with Docker Compose—separating the API, model, and database into lightweight services—and deployed it on a VPS with Nginx and SSL for production use. A Streamlit front-end provides an intuitive interface for interactive exploration of math concepts and proofs.
This, at the time, represented my first foray into working with open-weight LLMs, and LLMs in general. This was a humbling experience — I learned that LLMs are pretty far from plug-and-play on consumer hardware, and that inference/response quality expectations need to be managed when working with restrained resources.