Artificial Intelligence

AI Engineering Work

From agent orchestration frameworks to production LLM inference on custom GPU hardware - I build AI systems that work in the real world, not just in notebooks.

Hermes Agent Framework

Full AI agent orchestration with intent routing, MCP integration, entity extraction, and a 600K+ node knowledge graph. 4+ microservices with systemd health monitoring.

4 microservicesBM25 RouterMCP Server600K nodes

CodeGraphContext (CGC)

Code knowledge graph toolkit with LSP call resolution, Leiden community detection, Cypher queries, and hot-path risk analysis.

Multi-repoData flowCross-serviceClustering

AI Knowledge Router

Multi-system RAG routing that classifies queries and dispatches to GraphRAG, wiki-rag, or codebase-memory with confidence scoring.

4 backendsAuto routingConfidenceHybrid search

Mem0ai Memory Architecture

4-layer persistent memory with NER extraction, embedding quality monitoring, ChromaDB vector storage, and scene block synthesis.

NER >90%BGE embeddings1.8K entitiesScene blocks

Intel Arc B70 Inference

Production vLLM on 4x Intel Arc Pro B70 GPUs with NVFP4 quantization, tensor parallelism, and Docker toolboxes.

25-45 t/s gen5K-15K+ prefillNVFP4CI/CD toolboxes

Agentic Architectures Library

35 production-grade agentic AI patterns documented from real deployments with benchmarks and trade-off analysis.

35 patternsBenchmarksProductionOSS

My AI Philosophy

AI systems should be autonomous, auditable, and actually useful. Every system I build ships with verified outputs, comprehensive testing, and clean documentation. No black boxes, no hand-waving. Models are tools - not magic - and production-grade engineering is what makes them valuable.