Slack-first daemon that controls local coding agents with thread-based sessions. Features channel-to-repo mapping, automatic PR management, multi-agent support (Claude, Codex, Gemini), and real-time streaming output to Slack threads.

Built to learn fine-tuning principles and methods for LLMs utilizing PyTorch. Used RefCOCO dataset to fine-tune Qwen3-VL-2B-Instruct model. Achieved a 50%+ improvement in mean IoU compared to base model.

An MCP Server that connects you to the official SeatGeek Public Developer API. Written in Typescript, helps users connect to events in realtime on the SeatGeek platform utilizing tool-enabled LLMs.

Evaluation of popular vector database options, including Weaviate, Qdrant, pgvector, and more. Benchmarks retrieval speed, insertion speed, and accuracy using common embeddings. Adding more databases soon...

An LLM-Powered CLI tool written in Go that acts as your personal assistant. Includes a relay WebSocket server to connect computer and mobile device. Streams responses to terminal and mobile devices that can run Expo.

Scan ingredients with your camera and get personalized recipes using OpenAI. Trained ML model using YOLOv8 and public computer vision ready datasets. Stores to locally hosted DB to track recipes and scores.

Creates a personalized photography portfolio using an Instagram user's media. Previously serviced 12 paying users. Working on refactoring to allow more customizability and Stripe integrations.