AI-native medical billing
Voice agents that call payers, a denial agent over a 4.7M-row knowledge base, and an RCM core.
Nabbil KhanAn index of one operatorAs of Jul 01, 2026, 07:00 PM CT
Founder / Operator
My companies run on AI. Literally.
Not a metaphor. Four businesses whose daily operations are carried out by an autonomous AI fleet I designed and built, running in regulated industries around the clock.
4 companies/16 agents/0 failed services/49 days live
The companies
Four companies. One operator. One fleet.
On the record
Measured, not described.
0commits across
25 repositories0hourly snapshots,
no gap
25 repositories0hourly snapshots,
no gap
16 agents/6 active today/40 services, 0 failed/49 days uptime
Every figure is a command output on one always-on machine. As of Jul 01, 2026, 07:00 PM CT.
Selected work
Systems that do real work, not demos.
Self-improving training methodology
Agents that diagnose their own failures and promote only statistically defensible wins, overnight.
Frontier models on consumer hardware
Hand-written C and Metal engines running 397B and 671B open-source models on a 7.7-lb laptop.
Validatable pharma-compliance automation
57 minutes of manual closeout reduced to 3, built to pass a 21 CFR Part 11 audit, fully local.
Insights
One operator, many industries. What the real work reveals.
Not opinion. Patterns pulled from what actually shipped across healthcare, voice AI, model training, compliance, and more, written from inside the operation.
Writing
Field notes from inside the operation.
Latest signal · I don’t hire ‘AI people’ first anymore.
Speaking
I do not pitch with slides. I pitch with telemetry.
Talks on production multi-agent systems, AI in regulated industries, and building a company that runs on AI, backed by a real operation you can verify.
Topics + booking →Newsletter
Field notes from building and running AI in production. No hype.
