I build tools that I need for my work, most involve forensic analysis, machine learning, and ediscovery related tools. I've been working with ML\LLMs for over 15 years, in theoretical research and applied/production.
Adaptive vector search with self-correcting embeddings. Addresses semantic collapse in RAG systems through spectral chelation, dynamic dimension masking, and neural adaptation. Built with PyTorch and Qdrant.
Forensic-grade OCR platform for high-volume, mixed-format document processing. PaddleOCR with Tesseract fallback, language-aware routing across 176 languages, distributed queue processing via Celery/RabbitMQ, and evidentiary integrity preservation. GPU-accelerated with Docker deployment.
Autonomous AI agent orchestration engine with a local-first runtime, explicit governance, and extensible workflow automation. FastAPI backend with Ollama-powered LLM integration and a first-party control plane UI.
Scan and clean Twitter/X post history using local LLM analysis via Ollama with optional Claude second-pass review. Multilingual regex pre-filtering across 26 languages, interactive terminal review UI, and automated deletion through the Twitter API.
Structured framework for media-driven due diligence on recreational vehicles. AI-assisted photo and video analysis to identify defects, estimate repair costs, and build evidence-based negotiation packets.
Languages: Python, PowerShell, SQL
ML/AI: PyTorch, PaddleOCR, Tesseract, Sentence Transformers, Ollama, Claude API
Infrastructure: Docker, FastAPI, Django, Celery, RabbitMQ, Redis, PostgreSQL, Qdrant, OpenSearch
Domains: E-Discovery, Digital Forensics, Information Retrieval, NLP, Computer Vision, OCR