Working with me upstream

Most of my hands-on time goes into open source inference and routing: vLLM, vLLM Semantic Router (I'm in its Data Plane & Networking work group) and NVIDIA Switchyard. I learn a codebase by finding bugs and fixing them upstream, so every user gets the fix.

• Hit a bug in one of these projects? Tag @colinmcnamara on the issue or PR, or email it to me. I'm glad to help reproduce it, review a fix or co-author one.
• Working on /v1/messages? Making Anthropic's Messages API a first-class citizen alongside OpenAI's Chat Completions is my current focus, from the router to the engine.
• Want to start contributing? Ask me how I pick a first fix, write the repro and work with maintainers. I built gh-upstream-watch to keep track of it all.

The story of my first vLLM fix is The Router Was Right. What's merged so far is on GitHub and my about page.

Co-engineering at AHEAD

At AHEAD I work with enterprise and hyperscale customers on AI infrastructure. The best work happens when building at scale and contributing upstream feed each other:

1. 1. Build at scale

   Large deployments meet the edge cases first: real traffic, real clients, real failure modes.

   →

2. 2. Fix it upstream

   The fix lands in the open source project, not in a private fork your team has to carry.

   →

3. 3. Everyone benefits

   The next release ships the fix to every user, your next build included, and your engineers learn the code they depend on.

Then the cycle starts again from a better release. If your team is building AI infrastructure and wants to co-engineer this way, email me.

Speaking and workshops

Hard-won lessons from production, not theory, in three tracks. Linked titles are talks I've given, with slides.

Where I've spoken: a keynote, Shut Your Laptop & Go to Sleep!, at LinuxCon + ContainerCon Europe (Berlin, 2016), Open Networking Summit (2016), the OpenStack Design Summit (San Diego, 2012), the AI & Law Summit (Austin, 2026), the 39th Annual Legal Seminar on Ad Valorem Taxation (2026), and monthly at AIMUG.

AI infrastructure: inference and routing

• Mixture of models: cost, privacy and security at one control point with semantic routing
• Your harness is the other half of the model: failures that look like model problems and aren't
• The signal that can't lie: a biofeedback-driven, human-gated coding environment
• From Pregel to LangGraph: the evolution of graph processing

Law and regulated industries

• Hacking the law: what the machine reads that you can't: a CLE on the security and ethics of AI in legal practice
• AI governance: LLM-as-judge and continuous auditing for defensible AI
• From proof of concept to production: running agents in regulated industries (FINRA, FDA, NRC)

Business: rewriting the operating system

• Rewriting the business operating system with AI
• Agentic AI in the software development lifecycle
• Observable AI: why traceability matters for everyone
• Lessons from hyperscale: what Oracle taught me about AI operations

Every talk with slides is on the talks page.

Want me at your conference, company or class? Email me with the date, audience and topic.

Need a bio and headshot for the program? Here's my headshot and a bio you can use as is:

Colin McNamara is Field CTO at AHEAD in Austin, Texas, where he works with enterprise and hyperscale customers on AI infrastructure. Nights and weekends he contributes upstream to vLLM, vLLM Semantic Router (Data Plane & Networking work group) and NVIDIA Switchyard, working on mixtures of models and semantic routing. He founded AIMUG, the AI Middleware Users Group, 2,200+ members.