2024
Echo
Turns audio, video, or text lectures into clean notes, transcripts, and Q&A sets - cutting manual note-taking by roughly 60%.
Overview
Echo was built at the UCI MedTech Hackathon, where I led an agile team of nine developers. It starts from a simple premise - taking notes is now a thing of the past. Feed it a lecture as audio, video, or text, and it produces a transcript, structured notes, and a Q&A set you can study from, cutting manual note-taking by roughly 60%.
Like the clinical chatbot, it's local-first: models run through Ollama on your own machine (Llama 3.1 by default), and the backend is model-agnostic - point an environment variable at any model in the Ollama registry and Echo uses it. There's no API key and no per-token bill.
The architecture is a React single-page app talking to a Python Flask service that handles ingestion, transcription, and generation.
Highlights
- Built at the UCI MedTech Hackathon, leading an agile team of nine developers.
- Audio, video, or text in - transcript, structured notes, and Q&A out.
- Roughly 60% less manual note-taking.
- Bring-your-own-model: any Ollama model, configured via a single env variable.
- Runs fully locally - no API keys, no usage costs, lecture content stays on-device.