Projects
Things I build to think with — mostly agents, retrieval, and models that look at the Earth.
MACHINE LEARNING
Cashflow: Personal Finance AppA personal finance managment android app for tracking what I spend, own, and owe — savings runway, net worth, debt payoff, and FIRE projections. A statement-review pipeline reads a my bank PDF and optimizes for discrepancy detection, with on-device redaction before any LLM call, and a Pydantic AI agent resolves cryptic card descriptors into readable merchants. React + Capacitor (Android) frontend, FastAPI + Supabase backend.Mealsy: Agentic Nutrition CoachA personal agentic nutrition platform for logging meals in natural language via web chat or Telegram. A LangGraph agent extracts foods and looks up nutrients from the USDA database with Gemini fallbacks, while a Pydantic AI coach delivers personalized, data-grounded coaching from your eating history. Built on FastAPI, PostgreSQL + pgvector, Redis, and Celery.Poddersum: Agentic Podcast SummarizerAn autonomous agentic system built with LangGraph and Gemini 3.0 that automatically downloads, transcribes, and summarizes podcast episodes from gPodder. Features persistent multi-turn memory for interactive chat, real-time cost tracking, and autonomous background execution.Pal: Personal AI Wiki for ObsidianAn Obsidian plugin that turns your vault into an LLM-maintained personal knowledge base. Sources dropped into raw/ are automatically read, summarised, and cross-referenced by Gemini into an interlinked wiki, and a chat sidebar answers questions about your own knowledge with [[wikilink]] citations. Built in TypeScript on the Obsidian plugin API.