agent setup
Bottlenecks
patterns
Action items
Trusted by teams at
Frameworks I built — patterns I'll apply in sessions.
Orchestration patterns with tracing and failure guardrails. 40+ notebooks you can reuse in real flows.
Creator of LionAGI and former Autogen maintainer. I help teams build robust agent systems that actually work in production.
From debugging agent loops to architecting multi-agent orchestration, I bring practical solutions from real deployments. Whether you need MCP integration, failure handling strategies, or just a second opinion on your architecture — I've been there.
96K LOC pure Rust inference engine. SIMD kernels, Metal GPU, QuaRot Q4 quantization. The only working Q4 for Qwen3.5 GDN on Apple Silicon.
Rust MCP server with multi-tenant architecture, hybrid retrieval (HNSW + BM25), OAuth 2.1. Published on crates.io.
Formal verification in Lean4. 26K+ lines, 0 sorry. Proves core data structures via Charon/Aeneas extraction.
These frameworks encode the patterns I apply in sessions. Adoption optional.
Expert technical consultation with Ocean Li. Advisory sessions for agent systems architecture.
Quick assessment
Enough time to dig into a problem
Live problem solving
Full strategic session
All sessions prepaid via Calendly.
You bring the problem — I bring the patterns. We'll work through your specific challenge together, whether it's agent orchestration, MCP integration, or figuring out why your system isn't behaving as expected.
Multi-agent coordination, memory systems, capability scoping. The architecture that determines whether your agents compose or collapse.
Model routing, prompt structure, fine-tuning decisions. Where your token budget is going and whether it's going to the right place.
Rust and Python pipelines, embedding systems, vector search, MCP servers. The performance layer under your agent stack.
Build vs buy, team structure, prioritization. A second opinion grounded in having built the full stack myself.
Hooks, MCP servers, multi-agent workflows, CLAUDE.md design. Get your team shipping with Claude Code at the level I use it every day.
Send a brief description of what you're building and where you're stuck when you book. A diagram, repo link, or error log goes a long way. The more context I have beforehand, the deeper we go during the call.
Rust, Python, TypeScript. Agent orchestration (MCP, tool-use, multi-agent coordination), inference infrastructure, embedding/retrieval systems, formal verification. If you're building AI systems that need to work in production, we'll find common ground.
Yes. Everything discussed is confidential. I can sign an NDA before the session if your legal team requires it.
Yes — for teams building complex agent systems, I offer weekly retainers with code review, architecture check-ins, and async support. Book a 30-minute session first so we can scope it.
60-minute sessions include 30 days of async follow-up via email. For anything beyond that, book another session or ask about a retainer.
Pick a session and book a time. Share some context beforehand so we can make the most of it.
Book a session