Senior Technical Program Manager · AI-Native Platforms

I connect systems
that don't talk
to each other.

Technical, human, or both. 10+ years shipping complex programs across AI/ML, IoT, and hardware. This year it's Kestrel, a production AI platform I ship and operate solo, with evals in CI.

Vitalik Garan Frankfurt, Germany LinkedIn GitHub Email

The proof

Three artifacts, written for a 60-second read each. Numbers first, story second. Everything here is checkable against the repos.

Kestrel: an operated AI system

Solo-built production job-market platform. Scoring filter measured at 93.6% recall / 74.6% precision on a 277-item human-judged benchmark, evals in CI, AI cost per run cut $6.51 to $0.84. Plus an annotated repo tour.

Read the case study →

Ring radar: a senior-judgment one-pager

A radar camera from sensor research to 800k-2M units shipped. 10+ teams, 5 countries, 4 years. And the two calls that mattered more than any launch: a full platform rollback and a termination.

Read the one-pager →

Parallel invention: Kestrel vs career-ops

Two people built the same idea, published four days apart, without knowing of each other. Theirs has 62,000+ stars. Mine has 5, and measured evals. A teardown of what actually differed.

Read the teardown →

The short version

My brain works like a browser with 40 tabs open. I know a lot about a lot of things, and I'm not deep in any single one. That used to feel like a weakness. It isn't. It's the whole point of me.

What I actually do is take a service here, an integration there, a Slack feature nobody noticed, a Jira hook that no one bothered to set up, and make them talk to each other. It's not work. It's exploration.

Same with people. I connect teams and people and make them do their best work -- whether I'm "alone" in the hierarchy, like TPMs sometimes are, or part of a project or a team. What most needs connection, support, and direction is people, and the teams those people make up. My focus is people over tools or names.

People above tools. Roads above walls.

I function as a one-man army when I have to. Ownership means making things happen. But I'm only as good as the team, and I stand on behalf of the team and our decisions. The frontier work I did myself, I could only do thanks to the people who supported me: a team always on hand with expertise, kindness, thoughtful ideas, partnership.

Things I've shipped

The title says program manager. The work says builder.

Independent Feb 2026 · present
Builder & Operator, Kestrel

A deliberate choice after Wolt: go deep into the 2026 multi-agent AI stack instead of watching it from a deck. Working alone for half a year, I got promoted to Builder, Product Manager, QA, CFO, and CTO of myself. Kestrel is the AI job-search platform I run on myself: FastAPI + React, ten AI providers behind one interface, 4,600+ tests, on PyPI under AGPL. Open source in public.

  • Kestrel: scoring filter at 93.6% recall / 74.6% precision on a 277-item human-judged benchmark; AI cost per run cut $6.51 to $0.84.
  • screenpipe: 24 merged upstream PRs + a dev/test harness I dogfood 24/7 -- work that led to a direct offer from the founder.
  • More: web-agent-comparison, llm-safe-haven, awesome-llm-token-optimization, chronoclaude.
evals in CI · $6.51 → $0.84 per run
Wolt (DoorDash) Oct 2025 · Feb 2026
Technical Program Manager

Fed every API, transcript, doc, and private chat into an LLM system until it knew more about the program than any single person. 6 AI models, 4 enterprise APIs, one second brain that could synthesize status across the entire org in a single prompt. Volunteered to build it. Still love it.

  • Joined to build the AI-augmented program function from scratch, post-M&A, with no established tooling culture.
  • Managed the Salesforce CRM transformation across product, engineering, and business stakeholders.
6 AI + 4 APIs → 1 second brain
Amazon · Alexa AGI 2022 · 2025
TPM, ML & Speech Recognition

Ran continuous ASR training pipelines across French and Italian locales. Got researchers who want to experiment and engineers who need to ship into working alignment across 3 continents. Cut release SLA from multi-day to 1 business day.

  • Continuous ASR training and deployment across French and Italian locales -- research in Aachen and Bangalore, engineering in Seattle.
  • Drove an early LLM integration initiative alongside the ASR work, during the inflection year for large language models.
3 continents, 1-day SLA
Amazon · Ring 2018 · 2022
TPM, Sensor Research & Hardware R&D

Led Ring Ultra radar camera from sensor research to 800k-2M units shipped. 10+ teams, 5 countries, 4 years. ML, hardware, firmware, and software simultaneously. Shipped Motion Settings 2.0 to 40M+ devices. Managed full platform rollback after an accidental global deploy -- the response was the measure. The one-pager has the two hard calls.

  • Motion Settings 2.0 to 40M+ devices in controlled 1M/day waves; Person Detection from 30% to 46% on deployed hardware.
  • ANVIL compliance overhaul: 25 independently certified services, zero downtime, 5 certifications 30 weeks early.
  • Portfolio of 7 products fusing ML, hardware, firmware, and software.
800k-2M units shipped
CloudMade 2017 · 2018
Fleet & Field Program Manager

Built 5 modded VW demo cars running the company's stack natively. Reverse-engineered the CANBUS, pulled real sensor data. 200+ live demo drives. Sales engagement up 5x. Several $1M+ contracts directly attributed.

  • Cut demo prep from 1 month to 1 week by integrating the field team directly with R&D.
5x sales engagement

The paper version, with every date and bullet: the full CV as a page (print-friendly).

Things I believe

Status: incubating. Dictated, not polished -- some of these I can't formulate well yet, and that's fine.

People above tools, roads above walls

People above tools -- agile style. People are above tools, frameworks, names, companies, authorities.

Reason above frameworks

Reason above frameworks, expectations, and cool name-drops -- scrum and sprints and jargon. It all needs to be incubated.

Better work for a better society

Better work for a better society, for a better world, for people. Doing good work for an evil program or evil project or evil company which clearly harms us all, or harms other people somewhere else, is not a way to go.

Results in perception, process for teams

I cannot yet formulate it well, but: result in personal perception, process for teams. You want to be result-oriented as an individual -- when you look at a task or a problem or a project. But teams shouldn't be result-oriented as a communication thing; they need to build mechanisms (Amazon jargon) -- human mechanisms which work, which bring the process, which brings results.

The slide between scales

We all simplify complex systems to be able to understand them and talk about them and move forward -- as simple as possible, but not simpler. But then you have to complicate stuff when you're looking at real results: the map is not the territory, and a one-to-one map is no longer a map. You have to be able to do this mental slide between the scales -- if it's too simple, you lose the patterns; if it's too complex, you lose them anyway.

Secular but faithful

We operate in a space where -- let's do it metaphorically -- there is no God. But we still believe in the future, we still believe in good faith, we still believe in doing the right thing. And we still stand on the good side and fight the evil side, regardless of whether God exists or not.

Not a policeman but a maintainer

We all have to do things on our own, take ownership, hold each other accountable, and ourselves too. But we are an ecosystem, and we should function as one. Everyone has to maintain it, and I do everything I can to be not a policeman but a maintainer. Not a judge but a caring guardian of the ecosystem.

Trust is the fuel

I believe human relationship and human trust is the fuel that brings better products, a better society, a better future.

What I'm looking for

A team to join, and a team to build: its morale, its style. Senior Technical Program Manager, AI Program Lead, or Product Engineer at a company that actually wants to change how things work. Frankfurt, Germany -- remote-friendly.

I flourish in -1→1. Give me something broken or undefined and I'll find a way through it.

Senior Technical Program Manager AI Program Lead Product Engineer -1→1 AI-native Available now

Let's talk

Fastest way is email. Always up for a good conversation about building things that matter.

Email LinkedIn GitHub