Haider Abbas
Two production systems in regulated environments — one piloted inside a hospital, one taking live phone calls. Before that, two years underwriting a $2B venture portfolio at Silicon Valley Bank.
I find the workflow nobody has automated, size the business case, and ship the thing.
2,000
Documents automated
Two-month FormIQ pilot, Yale Health orthopedics
300
Admin hours recovered
Same pilot — about nine minutes per document
$2B
Portfolio underwritten
PE/VC fund banking clients at Silicon Valley Bank
80 hrs/wk
Manual work removed
Covenant-compliance automation shipped at SVB
Every number here was measured, not modeled. Where I have modeled something, I say so and show the assumptions it turns on.
Find the workflow that is still manual, learn it from the people doing it, build the part a machine should do, and keep a human in the path wherever being wrong is expensive.
Co-founder and tech lead
Oct 2024 – May 2026
AI intake for hospital fax, prior-authorization routing, and clinical review.
Live pilot at Yale Health, then final-stage procurement
2,000
documents automated
300
admin hours recovered
~9 min
saved per document
Founder
Aug 2025 – present
A production voice agent that takes live restaurant phone orders.
Public launch September 2026
Live
real inbound and outbound calls
Telnyx
full call lifecycle, own teardown
Fails closed
the model never owns state
Founder
Dec 2024 – Aug 2025
Voice and retrieval practice platform for consulting interviews.
Voice and retrieval, built during the MBA recruiting cycle
~500
cases in the library
Voice + RAG
live practice, structured scoring
We cleared security review. We cleared clinical workflow review. We reached final stage — and lost on vendor maturity. They wanted a company that would still exist in five years to support the thing. That is a reasonable requirement, and not one two people can satisfy inside a procurement cycle.
It was not a product decision, and no version of the product would have changed it. What I took from it: in enterprise healthcare the product is necessary and nowhere near sufficient — buyers are underwriting your continued existence as much as your software. I now ask about institutional credibility as early as I ask about technical fit.
Read the FormIQ case studyAug 2025 — present
Dohos
A production voice agent taking live restaurant phone orders. Full telephony path on Telnyx — call lifecycle, signature verification, idempotent event handling, teardown on every failure mode — over a Pipecat media pipeline with AssemblyAI, neural TTS and Vertex Gemini behind bounded fallbacks. Architected so the model proposes typed edits while deterministic code owns catalog identity, pricing and confirmation. Public launch September 2026.
Oct 2024 — May 2026
FormIQ
AI intake for hospital fax, prior-authorization routing and clinical review. Ran a two-month live pilot in Yale Health's orthopedics department — 2,000 documents automated, 300 administrative hours recovered — and reached final-stage enterprise procurement at an academic medical center as a two-person team, clearing security and clinical workflow review. The engagement ended on vendor-maturity requirements, not product fit.
Dec 2024 — Aug 2025
ConsultX
A voice and retrieval practice platform for consulting interviews, with roughly 500 cases, live mock interviews and structured scoring. Built the voice and retrieval foundation I later carried into Dohos.
Jan 2022 — Apr 2024
Silicon Valley Bank
Managed a roughly $2B portfolio of PE/VC fund banking clients — capital call lines, management company loans, NAV facilities, portfolio company loans — owning underwriting, credit approval, covenant compliance and default workouts. Shipped covenant-compliance automation with the enterprise team that removed about 80 team-hours a week of manual monitoring. Held founder and investor relationships through the 2023 collapse and the First Citizens acquisition.
Jul 2018 — Jan 2022
Pizza Boli's
Ran daily operations for a 20-person team — order takers, cooks and drivers — while completing a full-time undergraduate degree. Negotiated supplier changes that cut core food costs about 8%, and redesigned phone-order coverage with VOIP and offshore staffing. This is the problem Dohos now automates.
Roles where the hard part is the workflow rather than the model — sitting with a customer, working out what actually happens versus what they say happens, and shipping something that survives contact with operations.
Education