Work
What I do, how I lead, and what came of it.
A snapshot of the tools and the way I work, then a few engagements that show it in practice.
Skills & tooling at a glance
The short version, if you'd rather not read the case studies — what I do, the tools I reach for, and how I work.
Tech stack
Languages & frameworks
Data
AI & LLM
Optimization & modeling
Infrastructure
How I work
How I build
I'm at my best taking something from zero to ten. When the problem is still vague and the path isn't drawn. I design a way to answer the question, ship a real working thing, and stay honest about what the results actually mean.
How I lead
Direction, the first hires, and an operating rhythm that ships. I've built a data/AI team from zero to four and run it in a Shape Up cadence: problems shaped with leadership, fixed appetites, and bets that move to production or get dropped without ceremony.
How I collaborate
Embedded, not over the wall. I work inside your team: building the production feature alongside your engineers in a tiger-team model; then hand off something that runs without me.
How I treat people
The non-negotiables: respect first, trust by default, radical candor when it counts. I enable people rather than manage them, keep a “find a way” bias, and build teams that are safe and welcoming — including for queer and neurodivergent folks, which I care about as someone who’s both part of and building for those communities.
Case studies

Soda · Intelligence & Automation Products Lead · 2021–2026
Founding Soda's AI data-observability research team and product range
Team built: 0 → 4 · Flagship AI features shipped: 5 · Commercial impact: Top OSS→paid lever
Read case study →
Navan · Senior Data Scientist · 2019–2021
Shaping Navan's data team into a central, product-facing resource
Content-discovery time: 30s+ → <10s · Squads led: 2–4 DS, multiple lines · Product lines: Several, in parallel
Read case study →
TravelBird · Data Scientist · 2016–2018
A one-person data-science function at TravelBird
Attribution model: +€5M/yr net profit · Catalogue trimmed: −40%, revenue flat · Cold-start solved: ~40% of catalogue
Read case study →