Small, practical experiments across audio fingerprinting, traffic clustering, autonomous reporting, and local multi-agent systems. Each project starts with messy input and ends with something a person can use.
What signal is hidden in the data? These projects explore the pipeline from raw information to useful output: extract the pattern, test the model, and make the result legible.
Identify a song from a short audio sample by transforming sound into a spectrogram, finding its strongest peaks, and matching the resulting fingerprints against a main-song database.
A smart traffic-management simulation that uses random mobile-phone “dots” to represent passengers, then clusters nearby signals to estimate where vehicles are moving.
The model is only one part of the work. Each experiment is shaped around how a person will read, trust, or act on the result.
Collect signals from sound, location-like dots, feeds, or conversations.
Extract features, cluster patterns, summarize evidence, or coordinate agents.
Return a match, estimate, report, or recommendation that can move work forward.
A scheduled agent that collects news from global, sports, and local sources; summarizes it with Google Gemini; tracks AAPL, NVDA, and GOOG with yfinance; and sends a rich HTML report by email.
Built with Google Antigravity and deployed on Google Cloud with a scheduler. View the project on GitHub.
Three specialized Gemma 3 4B agents—Maverick, Guardian, and Mediator—debate a question locally to expose upside, downside, and raw facts.
View the CouncilRead the detailed Council architecture or try the experimental Maverick interface.
Gemma CouncilMaverick ChatOpen the notebooks above, or browse the source links inside each project.