// projects
Open-source tools and experiments. Each one started with a problem I couldn't stop thinking about.
AI-powered codebase auditor. One command analyzes your entire repo for security vulnerabilities using multi-agent architecture with specialized security domain agents.
Uses Claude to orchestrate multiple specialized agents — authentication, injection, crypto, configuration, file handling, and data exposure — that each analyze the codebase through their domain lens. Results are deduplicated and ranked by severity.
Convert PDFs to clean markdown using LLMs with vision capabilities. No traditional OCR — just send pages to a vision model and get structured output.
Traditional OCR struggles with mixed layouts, tables, and mathematical notation. This tool renders PDF pages as images and uses vision-capable LLMs to produce clean, properly formatted markdown. Works especially well for academic papers and technical docs.
Python asyncio library for concurrent and time-based rate limiting. Clean API for controlling request throughput in async workflows.
Provides both concurrency limiting (max N concurrent tasks) and time-based rate limiting (max N requests per second). Composable with existing asyncio patterns. Used in production for API client rate limiting.
Natural voice conversation with LLMs. Contributed to the open-source framework for real-time, full-duplex voice chat with language models.