The firm
Built for the part
after the demo.
AI work gets real when it has to survive the workflow: the approvals, edge cases, handoffs, controls, and people who own the outcome. Lumi helps teams turn that messy middle into working software. We build governed production agents, embed with the teams responsible for launch, and use Kestrel Suite to make the work visible, testable, and repeatable.
What Lumi is
- A practical AI delivery partner
- A builder of governed production agents
- A team that starts with the workflow, not the model
- The people behind Kestrel Suite: delivery infrastructure for repeatable, observable agent work
What Lumi is not
- Not a slideware AI strategy shop
- Not a chatbot vendor chasing novelty
- Not a staffing firm wearing AI language
- Not a platform company that disappears before production
How we work
Clarity
We say what is known, what is assumed, what is risky, and what it will take. No inflated transformation language. No mystery process. Every serious claim should be tied to evidence.
Ingenuity
The useful answer is not always the fashionable one. Sometimes the win is an agent. Sometimes it is a better workflow, a sharper approval path, a cleaner handoff, or a test harness that finally shows what is breaking.
Trust
Controls are part of the build, not a cleanup step. Permissions, reviews, traces, evals, and operating evidence belong in the system before the work is trusted with real outcomes.
Progress
We do not treat every project like a one-off experiment. Each launched workflow should leave behind reusable structure: maps, tests, patterns, integrations, and proof the next workflow can build on.
The bench
Amar Naga
Founder & CEO
Builds the practice. Chief executive and AI strategist with twenty years leading enterprise delivery, from global consultancies to founding Lumi. Forty Under 40 winner, Elevate Texas Award recipient, and focused on one problem: AI that survives contact with production.
LinkedInGreg Asher
CTO · Co-founder
Builds the platform. Greg created Kestrel around a hard-earned belief: production agents need evidence, controls, replay, and handoff from the beginning. Otherwise the pilot never becomes a system people can trust.
LinkedIn