Reducing the effort of food logging
This capstone explored a simple input: describe a meal aloud. The prototype converts speech into a structured food diary, connecting transcription, food extraction, and food matching.
Leading the product and the pipeline
As team lead and data scientist, I developed the product roadmap and coordinated development tasks. The team built a Streamlit application with Docker and FastAPI, using Whisper for transcription, Food-Based-BERT for extraction, Sentence-BERT for embeddings, and FAISS for matching.
From coursework to an internal beta
The project culminated in an internal beta at the company. It brought together a user-facing workflow and the model pipeline behind it.