AI / LLM Infrastructure Design
Production architecture for LLM systems: orchestration and routing layers, retrieval, evaluation harnesses, low-latency inference, and reliability frameworks that hold up under real load.
I'm Henry Duong — technical founder, chief scientist, board director, and fractional CTO. MD-trained, Silicon Valley-built. I design, review, and de-risk AI/LLM infrastructure and agentic systems — the kind where multiple models collaborate, reason, and act in real time.
$ I make and break things.
I've spent 15+ years building and breaking AI systems in the real world — not in slide decks. I co-founded Sully.ai and scaled its engineering and core technology from a ~$35M seed-stage company to ~$250M post–Series A, architecting a multi-agent clinical AI platform that scribes, reasons, and acts inside live EHR systems. The companies I've built have drawn backing from people and institutions who don't write checks lightly — Mayo Clinic, Zoom founder Eric Yuan, and Rocket Money.
Before that I took Intellabridge from pre-seed to a public listing (CSE: KASH), grew the org from 10 to 70+ engineers across continents, led GIBLIB's Mayo-Clinic-backed surgical AI platform, and built DiagnosisAI — deployed to Special Operations under a DARPA contract and partnered with NASA Ames.
I'm an MD who learned to ship production software, which means I translate fluently between the clinic, the boardroom, and the codebase. I hold a granted US patent on multi-agent clinical decision systems and published Second Opinion Matters on consensus model ensembles. I've been a guest lecturer at UC Berkeley, course advisor at Stanford, and an AWS CTO Fellow.
Lab555 is named after the 555 timer — the first chip I ever wired, at age 10. It still does one thing perfectly: it keeps time, reliably, forever. That's the bar I hold for the systems I help you build.
Sector-agnostic. I've shipped in healthcare, finance, legal, enterprise, consumer, and defense-adjacent environments. Pick the depth you need — strategy, architecture, or hands-on.
Production architecture for LLM systems: orchestration and routing layers, retrieval, evaluation harnesses, low-latency inference, and reliability frameworks that hold up under real load.
Designing systems where specialized agents collaborate, reason, and act — consensus and ensemble reasoning, tool use, and human-in-the-loop guardrails. The space my patent is written in.
Independent audits of models, code, and IP: readiness reviews, failure-mode analysis, eval design, and a clear-eyed read on what's real versus what's a demo.
Diligence for funds and acquirers — I was the technical authority for a $500M fund. Architecture, team, scalability, and risk, distilled into decisions you can defend.
Wiring AI into the workflows it has to live in — EHRs, enterprise systems, real-time pipelines, and APIs — so models produce structured output people actually use.
0→1 and 0→IPO scaling, org design, technical roadmaps, and fundraising support. I've grown teams from 10 to 70+ across 8 countries — and I take on board director and advisory seats where I can move the needle.
Clinical AI, EHR workflows, surgical video, telemedicine.
Blockchain infrastructure, payments, a public listing.
Regulated data, privacy, security-first architecture.
Distributed systems, MLOps, global engineering orgs.
Cross-platform apps with 100K+ users; voice & mobile.
DARPA-deployed decision support; NASA-affiliated R&D.
Built and scaled engineering from early launch through deployment — ~$35M seed to ~$250M post–Series A. Architected a multi-agent clinical AI platform and the "Sully Intelligence" LLM orchestration layer; shipped the first agentic AI employee, scaling to six. Patent + publication.
Technical authority for a $500M fund (ex–Deutsche Bank leadership). Led all technology diligence and evaluation decisions and advised portfolio companies as fractional CTO.
Led the technical vision for a Mayo-Clinic-backed AI medical video + VR surgical platform. Built NLP-driven ingestion and DNN/CNN/RNN video understanding, turning vast unstructured footage into structured learning systems.
Took the company from pre-seed to a public listing (CSE: KASH / OTCQB: KASHF). Scaled engineering 10 → 70+ globally and architected blockchain-based financial infrastructure.
Built an NLP medical Q&A system deployed to Special Operations via a DARPA contract and partnered with NASA Ames. Cross-platform apps and a voice skill reaching 100K+ users.
"Providing insights to a doctor from a consensus of multiple AI specialists analyzing conversation." A multi-agent AI architecture for real-time clinical decision support — with two additional patents pending.
Tell me what you're building or stuck on. Whether it's an architecture review, an agentic pipeline, diligence on a deal, or a fractional CTO seat — I'll tell you straight whether I can help and how.