Open to roles · Product Analyst · TPM · Product Manager · remote-first
Data Product Manager · San Francisco, CA

The AI-native PM
who ships.

I run AI agents like an engineering team: 100+ production PRs shipped solo, every decision grounded in the data. Most recently I conceived Telos, an AV safety platform, and built it end to end. It's now in the hands of NHTSA and the California DMV.

Book a 15-minute intro ↗ See the work ↓

Every claim on this page links to something you can open: a live product, a repo, or a case study.

Telos platform: a critical near-miss in Martinez, CA surfaced out of 650 detected events, scored by physics-based risk with contributing factors fused into one view
Fig. 01 · Telos: 650 road-safety events scored by physics-based risk; one critical near-miss surfaced with its causes fused into one view. Case study →
NHTSA + DMV
Regulators using work I shipped
100+
Solo production PRs at AMG
−78%
False alarms cut by tuning detection
5
Public projects, numbers checkable

The flagship

01 · Telos
Delivered · In regulators' hands Sep 2025 – Present · Advanced Mobility Group · Walnut Creek, CA

The AV safety review that took 4–6 months now takes a minute.

Telos: conceived, built, and priced by me, solo, from an empty repo

In a working session with the DMV, I saw that the agencies responsible for AVs had no real way to see the edge cases: the moments where autonomous driving goes wrong on real streets. I built Telos to close that gap: every relevant data source fused into one plain-English view, so the people overseeing AVs can see what happened, understand why, and act the same day.

  • Delivered to NHTSA and the California DMV to explore and cluster AV safety events, the exact users that DMV session said were flying blind.
  • Three cities asked to deploy it after demos to their county transportation authority. They wanted the same risk signal as decision support for their own safety infrastructure, making everyday driving safer, not just AVs. The company has since moved to commercialize it in-house.
  • Physics-based risk scoring (time-to-collision, RSS) with a local-LLM narration layer. Cut false alarms 78%, turning a noisy telemetry feed into a decision-grade signal.
  • Owned the product end to end: PRD, roadmap, pricing and buyer mapping, KPIs, and the full FastAPI/Next.js implementation.
Fig. 02 · Highlight reel: fleet overview, plain-English event narrative, dashcam evidence with telemetry, hotspot clustering.
Read the full case study → Skip to outcomes →
1 min
To a defensible safety diagnosis
650
Events scored in one deployment
3
Cities asked to deploy after demos
100+
Production PRs, solo, full-stack

Don't take my word for it. Open it.

One domain, safety-critical transportation, attacked from five angles. Code you can read, numbers you can check, demos you can click.

Full build log →
B-01 · Data

SGO-Audit Live

Audits 4,589 NHTSA records of the nation's public AV crash dataset: heuristics, a disagreement model, and a local LLM, framed as review candidates, never verdicts. 0.912 precision at the chosen operating point.

Showcase ↗Repo ↗
B-02 · Perception

VRU-Detect Live

Turns a perception failure mode, detectors missing the pedestrians who matter most, into an auditable benchmark. Precision 0.396 → 0.685 baseline to fine-tuned, with a model card stating where it still fails.

Case study ↗Repo ↗
B-03 · Triage

Safety Event Triage Agent Live

Regulators give operators 24 hours to report a qualifying event. A config-driven agent classifies severity and drafts the reviewer summary, fully local by constraint, measured against 60 hand-labeled ground-truth rows.

Demo ↗Repo ↗
B-04 · Product

ZeroPath Live product

Cities wait 4–6 months for an intersection study that mostly assembles data they already have. Crash data in, PE-reviewable draft report out, in under 30 seconds. Built as a real product with a written GTM playbook.

crossingzero.io ↗
B-05 · Process

Agentic SDLC Toolkit Open source

The published engineering discipline behind all of it: six staged gates, each checked by an independent adversarial reviewer agent. The author never signs off on its own output. The process behind 100+ solo production PRs.

Repo ↗

One builder. Five seats.

Titles vary; the work is the same. Hiring for any of these? The evidence below is already yours to check.

02 · Where I fit
Seat 01
Data Product Manager

Ships data products end to end. Telos fuses telemetry, camera, weather, and road data into a decision-grade risk signal; Omnisync's hybrid-search databases served millions of documents.

Primary lane
Seat 02
Product Analyst

SQL-native. Traced researcher churn to model accuracy at Eclipsebio and helped close the gap (+8%); picked Telos's detection thresholds from the data, cutting false alarms 78%.

Analysis first
Seat 03
Technical PM

Owns the spec and the stack: PRDs, pricing, buyer mapping, plus the FastAPI and Next.js implementation behind them. Nothing gets promised that can't be built.

Spec + stack
Seat 04
Product Engineer

100+ solo production PRs at AMG, a Stripe marketplace built end to end at Plate Plan, 1,300+ automated tests at Dexcom. AI agents as the team, production quality as the bar.

Ships daily
Seat 05
Program Manager

Ran six concurrent deliveries across 30+ contributors at CSES; published the six-gate agentic SDLC that keeps parallel work disciplined, with an adversarial review at every gate.

Parallel delivery

Track record

03 · 2022 – present
2025 – now
Advanced Mobility GroupTechnical Product Manager / Systems Engineer

Conceived and built Telos, an AV safety platform, 0→1 solo: physics-based risk scoring, local-LLM incident narration, pricing and buyer mapping. Delivered to NHTSA and the California DMV.

−78% false alarms
2024 – 25
Plate PlanFounder & Product Manager

Built a two-sided meal-prep marketplace solo (Next.js, Postgres, Stripe): 10 chef partners, ~50 paying customers. Funded the build with freelance PM consulting for 3 startups, mentored by an Amazon PM.

10 supply partners
2024
EclipsebioProduct & Data Science Intern

Traced researcher churn to model accuracy through benchmarking analysis, partnered with data science to close the gap, and shipped a streamlined lab-to-analysis protocol.

+8% model accuracy
2023
DexcomSoftware Engineer, Intern

Wrote 1,300+ automated tests for 70 manufacturing-line scripts on FDA-validated software; fixed 200+ bugs with Python and SQL.

1.5× production speed
2023
OmnisyncSoftware Engineer & AI/ML Intern

Built ETL pipelines (Python, AWS Lambda) ingesting millions of documents; stood up Weaviate hybrid-search databases and APIs; managed RDS schemas for clients including the Department of Energy.

Millions of docs
2023 – 24
CSES DevFounder · Student-Led Software Studio, UCSD

Founded and ran a 30+ person studio delivering full-stack products for nonprofits and student orgs: six concurrent deliveries across PMs, engineers, and designers.

6 concurrent builds
2022
Packed (DormIt)Founding Product Manager

Led MVP to launch; raised onboarding completion 15% through 30+ user interviews and A/B testing; cut churn 20% with a gamified loyalty program built with 12+ engineers and designers.

$10K+ MRR

The toolkit

04 · Daily drivers
Data & Analytics
  • SQL: PostgreSQL, MySQL
  • Python: pandas, NumPy, scikit-learn
  • Tableau · Excel
  • ETL pipelines · A/B testing
Engineering
  • FastAPI · REST APIs
  • TypeScript: Next.js, React
  • AWS: Lambda, S3, RDS
  • Weaviate · Git
AI / ML
  • Claude Code: agentic SDLC
  • Ollama: local LLMs
  • OpenAI APIs · Cursor
  • YOLO fine-tuning
Product
  • 0→1 strategy · PRDs
  • User research · KPI definition
  • Pricing · GTM
  • Roadmapping

A bit about me

05 · Background

I've never fit neatly in one box, and that turned out to be the whole advantage.

I've always run a little ahead of schedule: high-school valedictorian at 16, UCSD graduate at 20 in Bioinformatics with a Computer Science minor, a degree that mixes biology, computer science, and data science. I'm still learning something new every week. I just don't like waiting to start.

At UCSD I became the only non-CS-major president of the CSE Society, then grew it from one of the smaller clubs on campus into its largest engineering organization, 400+ members strong. I wasn't the default candidate; I got there by building things people wanted to join. Under it I founded three organizations: CSES Dev, a 30+ person studio shipping software for real clients; CSES Open Source, which grew two projects to 30+ contributors; and CSES Innovate, an AI startup incubator that built a 20-school AI alliance and took six student startups from zero to MVP.

What I'd point to as genuinely rare is the breadth. I ship production software (100+ solo PRs at AMG, a Stripe marketplace end to end). I do real data science and AI work (model benchmarking, YOLO fine-tuning, local-LLM pipelines). I'm SQL-native, so the analysis that finds the problem and the fix that ships come from the same person. And it all runs through a product lens: users, pricing, and whether it's worth building at all. That combination usually lives in four different people, and the seams between them are exactly where I'm most useful. The same follow-through shows up off the clock, in a 1,000-lb club total.

Right now I'm looking for a team where that breadth compounds: product analyst, technical program manager, or product manager work, somewhere AI-native building is an advantage rather than an anomaly. If that sounds like your team, I'd like to talk.

Next step

The fastest way to evaluate me is a conversation.

Fifteen minutes. I'll walk you through Telos: the product calls, the analysis, what I'd do differently. You'll know whether I fit your team.

Book a 15-minute intro ↗ khouryn77@gmail.com
LinkedIn/in/nabilhannakhoury ↗
GitHub/nktarasios ↗
LocationSan Francisco, CA · remote-first
Open toProduct Analyst · Technical Program Manager · Product Manager