Project Detail

LecturePilot

An AI study tool that chats with your actual course materials.

LecturePilot lets students upload lecture slides, notes, and recordings, then ask questions against those materials through a source-grounded AI chat interface. The goal was not just to make another chatbot, but to make studying large amounts of course content feel more searchable, organized, and interactive.

Repository Live Demo

Problem

Students often have lecture materials scattered across slides, notes, and recordings with no good way to query them as a single study surface.
General-purpose chat tools are not reliable enough for course study unless answers are grounded in the actual uploaded material.

What I Built

A full-stack AI study platform where users upload lecture content and interact with it through chat, flashcards, and quiz workflows.
A retrieval-first UX that keeps the model tied to indexed source material rather than generic responses.

Technical Architecture

Next.js frontend paired with a FastAPI backend for ingestion, retrieval, and generation workflows.
PostgreSQL and pgvector for document storage, chunk retrieval, and semantic similarity search.
OpenAI embeddings and LLM calls layered on top of retrieval to produce source-grounded answers.

Key Challenges

Designing an ingestion pipeline that could process multiple material types asynchronously.
Balancing retrieval quality, response usefulness, and grounding so generated answers stayed tied to course context.

Impact / Results

Shows the ability to ship a complete AI product with practical retrieval constraints and clear user value.
Represents hands-on work at the intersection of full-stack engineering, data pipelines, and applied LLM workflows.