I build AI automation pipelines, agentic systems, and the orchestration infrastructure that makes them work in production.

10+ years of software engineering, from large-scale data processing pipelines at Netflix (Eyeline Studios) to designing multi-agent architectures and self-hosted AI orchestration stacks. I build the systems that let small teams ship like large ones.

Built automation infrastructure for Eyeline Studios' production pipelines at Netflix, processing terabytes across 100+ concurrent data sources with async orchestration and distributed queues.

• AI infrastructure: self-hosted orchestration stacks with Airflow, Qdrant, and full observability pipelines
• Multi-agent orchestration: autonomous AI agents with LLM routing, persistent memory, and webhook-driven automation
• Code intelligence: giving AI agents structural understanding of codebases via knowledge graphs and hybrid search

I built and operate a multi-agent AI system that functions as an autonomous software team. I create GitHub issues, and the system handles PRD writing, implementation, CI fixes, and review responses end to end without intervention.

The system uses 3-tier LLM routing: a dispatcher (Claude Opus) analyzes each request, selects relevant skills, and spawns workers (Claude Haiku) for parallel execution. A GitHub App with org-wide webhook delivery via Cloudflare Tunnel and cryptographic signature verification enables automated CI failure response, PR review handling, and issue triage.

• Agent presets and custom automation skills for domain-specific AI workflows
• Worker isolation in git worktrees for parallel autonomous development across multiple repositories
• Persistent memory systems and state tracking across agent sessions
• Full PR lifecycle management: CI failure to auto-fix, review feedback to auto-revision
• Prompt iteration and output validation via Langfuse for continuous agent quality improvement
• 7,300+ automated task commits since January 2025

Stack: Rust · Python · GitHub App (multi-repo) · Cloudflare Tunnel

Production infrastructure for AI workflow automation, running 12+ Docker services with resource-budgeted orchestration.

• Apache Airflow (Scheduler + CeleryExecutor) for DAG-based AI task scheduling, triggering agent workflows via REST API
• Qdrant vector database with GPU acceleration (CUDA) for embedding workloads and semantic search
• Full observability pipeline: OpenTelemetry Collector, Prometheus, Loki, Grafana dashboards
• RBAC per data domain for AI agent access control and GDPR sanitization on PII-bearing results
• PostgreSQL + Redis data layer; Traefik reverse proxy with automated TLS
• Tailscale mesh VPN for secure remote access, zero public ports for admin interfaces

Stack: Docker Compose · Apache Airflow · Qdrant · PostgreSQL · Redis · Prometheus · Loki · Grafana · OpenTelemetry · Traefik · Tailscale

A code intelligence engine in Rust that parses source code into a queryable knowledge graph, so AI agents understand architecture before they write a single line. Published on crates.io.

Instead of letting agents grep through files and guess at structure, Myceliums gives them a structured graph of symbols, call chains, functional modules, and execution flows, all queryable through MCP, a CLI, or a custom Cypher dialect. One search call instead of fifty.

• 6-crate Rust workspace covering core, storage, MCP server, Cypher engine, HTTP server, and CLI
• Parses 23 languages via tree-sitter with two-pass symbol resolution
• Hybrid search: BM25 + vector embeddings (all-MiniLM-L6-v2 via fastembed) + Reciprocal Rank Fusion, with optional cross-encoder reranking (BAAI/bge-reranker-base) and IVF-PQ indexing for large repositories
• Impact analysis: parses git diffs, then BFS through the call graph to find blast radius
• 12 MCP tools with one-command setup for 12 platforms

Stack: Rust · tree-sitter · LanceDB · fastembed · Cypher