A Daniel Pourhadi Foundation Model · v4.0

dan-4

The most capable Dan to date.

Principal / Lead engineer. Sets technical direction with executives, keeps design and product rowing with engineering, and turns “what the business actually wants” into a system that ships. dan3 promised to change the world. dan-4 shipped to production.

move your cursor · the network responds
01

Capabilities

core / 4
Architecture

Systems Design

Co-architected an entire platform from an empty repo — Kotlin Multiplatform, Micronaut microservices, GraphQL. The kind of foundation that's invisible when it works and named in the postmortem when it doesn't.

Leadership

Technical Direction

Partners with executives on strategy, keeps design and product aligned with engineering reality, and translates “the business needs synergy” into a roadmap that survives contact with production. Grows engineers; owns the hard calls.

Inference

Agent Engineering

Builds complex AI agents from the ground up: a patient-facing assistant wired into live FHIR medical records, and a real-time voice agent that runs the support phone line. It does AI. It will not bolt a chatbot onto your homepage to make the board feel current.

Alignment

Allegedly Aligned

Fine-tuned via RLHF, most of it stakeholder feedback. Declines harmful instructions, scope creep, and any meeting that should have been an email.

02

Benchmarks

DanEval v4

Evaluated against prior checkpoints and the median engineer. Numbers are accurate to within ±100% and were completely cherry-picked.

DanEval is authored by Dan, peer-reviewed by Dan, and reproduces reliably on Dan's machine.

03

Model Card

spec

Specifications

parameters~13B (years)
rolePrincipal / Lead
specialtyArchitecture
baseChicago, IL
deploymentRemote-ready
licenseHire-only

Training Corpus

  • 2021–now   K Health — co-architected the next-gen platform and led the team building it alongside design and product. Built K's most complex AI agents — patient-facing and voice — from the ground up.
  • 2020–21   Elevance / Anthem — Principal AI Engineer; shipped a symptom picker to 40M members across two large eng orgs.
  • 2017–20   Rocket Wagon — full-stack consultancy; ML for a wine distributor, battery-life models with Duracell.
  • 2009–12   Chicago City Council — Legislative Aide. The hardest distributed systems run on people.

Demonstrated Abilities

  • Server-driven native UI — a Kotlin Multiplatform engine that updates live apps with no App Store release.
  • PatientGPT — a patient-facing AI agent grounded in each user's live FHIR records, built from the ground up.
  • AI voice agent — runs the support phone line: answers questions, books, reschedules and cancels appointments at real-time-voice latency.
  • ML Trading Platform — LSTM, GRU & transformer models on index futures.
  • OnCue — top-10 paid music app; first crossfading queue player on the App Store.
04

Playground

temp 0.7
dan-4 · system: “be helpful, stay deadpan”
>

dan-4 may produce confident outputs that later require a hotfix. Not certified for safety-critical systems or standups.

Limited availability

dan-4 is currently deployed.

Inference capacity is fully allocated to an existing production workload. dan-4 maintains a short waitlist and will reprioritize for a sufficiently interesting problem.

Compensation negotiable · a system that's on fire moves you up the queue.