Code & Projects

What if you could search a whale call? MobyGrep listens to live underwater microphones (hydrophones) and runs the audio through Perch 2.0, a pretrained bioacoustics model, to turn each sound into a numeric fingerprint. A classifier trained on expert-labeled Orcasound recordings decides whether it's a whale call, and every call becomes searchable by similarity in Postgres with pgvector. Accuracy is measured against OrcaHello, the existing open-source orca detector.

Shevinu's Digest is an automated Python pipeline that distills 500+ daily tech stories into a ranked 9-item read using a four-stage relevance filter (model2vec embeddings, trafilatura extraction) followed by an LLM pass using 3 models.