ThinkStream Labs is the fractional practice of Ken Romero — a veteran systems-oriented technologist working with startups and small teams on AI infrastructure, edge systems, technical architecture, and the prototyping layer that holds the rest together.
Most teams I work with have one or two strong engineers but no one with the breadth to make a sane decision about which model, on what hardware, behind what API, with what governance — let alone how it all fits with the systems they already have.
That is the gap I sit in. I do the architectural thinking and the hands-on prototyping in the same week, for teams that need senior judgment more than they need another full-time hire.
Engagements are scoped around the outcome you need, not a generic service category. Each starts with a 20-minute call to figure out which shape fits.
Ongoing engagement, roughly one day per week. I become your senior systems brain for architecture, AI infrastructure, and integration decisions. Best for teams in motion that can't pause to hire.
For non-technical founders and early-stage teams. Covers the leadership function — roadmap, hiring input, technical strategy, vendor selection — without committing to a full-time leader.
A scoped engagement focused on AI workflow design, local-first setup, cost predictability, and the operational layer underneath. Typically 2–6 weeks, fixed scope.
Fixed-scope build of a working prototype, end to end. Best for teams who need to validate a technical direction before committing engineering resources.
A glimpse at the kind of problems I take on. Selected work from independent R&D and recent engagements.
Portable trust infrastructure for autonomous agent systems. Hash-chained audit trails, cryptographic identity, scoped capability grants, WASM-extensible policy engine, distributed consensus. Deep coverage across NIST AI RMF, OWASP LLM Top 10, OWASP Agentic Top 10, MITRE ATLAS, and the MCP Top 10.
A working map of agent and AI system security across the major frameworks (NIST AI RMF, OWASP LLM & Agentic, MITRE ATLAS, MCP Top 10), with notes on where current primitives fall short and where structural defenses change the math. Coming soon to /writing.
Working demonstrations of multi-agent governance in motion: hash-chained audit trails, three-party trust establishment, guard evaluation, delegation chains, and a 3D simulation of a fleet operating under policy. Coming soon to /demos.
The interesting technical problems live where domains overlap. That's where I'm most effective — and where most fractional engagements end up.
Book a 20-minute intro call. I'll listen, ask a few questions, and tell you honestly whether I think I'm the right fit. If I'm not, I'll often know someone who is.