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.
PIPELINE REPO for high-volume GPU/CPU OCR, mixed-format document processing, and other downstream enrichments. 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.
A unified Model Context Protocol (MCP) server that aggregates, indexes, and dynamically serves over 200+ specialized Agent Skills, Prompts, and Coding Workflows to AI assistants.
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.