Origin
UPLABS was founded in 2022 by Ahmet Alpat, an electrical engineer who'd spent the previous decade building AI systems in industry — and watching the same pattern play out: good research that died on the way to product, and good operators who couldn't access the research.
The original thesis was simple: be the lab we wished had existed when we were inside companies trying to ship hard things. Build a research practice that bills like a consultancy, ships like a product team, and converts the strongest projects into venture-stage companies.
The lab opened in Istanbul. The marine ops office opened in Lisbon when Poseidon was spun out in 2023. The cohort grew from one venture to six in four years. The thesis has not changed.
Operating thesis
We believe four things, strongly enough to bet the firm on them.
Research has a half-life
A 2024 paper is worth ten 2018 papers. We staff for currency, not seniority alone.
The same hands
Whoever scoped it ships it. Whoever shipped it documents it. No bait-and-switch.
One platform across the cohort
Inference runtime, MLOps, and eval harness reused from venture to venture. The seventh ships faster than the first.
Patient by design
Long retainers. Patient board seats. We don't churn relationships and we don't push exits.
How we work
Three things make the lab feel different from the inside, and probably from the outside too.
Bi-weekly demo or it didn't happen
Every engagement runs on a fortnightly demo cadence. Not a status update — a working artefact, in the actual repo, with the actual data. If a fortnight goes by without something to demo, the lead has to write a kill memo explaining why.
Senior squads, no pyramid
The median engineer at UPLABS has nine years of experience. We don't run a delivery pyramid because we don't believe juniors can do the work, supervised or not. They can, eventually. But our pricing model assumes everyone in the room is a peer.
Off-ramps in every contract
Every retainer has a quarterly no-fault off-ramp. The studio SAFE has a 24-month buyback clause. We'd rather lose revenue than carry a doomed engagement, and we'd rather lose a venture than block a founder.
What we don't do
The shape of a firm is defined by what it refuses. Here's the refusal list.
- No staff augmentation. If you need bodies, hire them. We sell shipped artefacts, not seats.
- No "AI strategy" workshops. If your strategy comes from a slide deck, it's already wrong.
- No predatory defence work. Defensive sensing & perception, yes. Lethal autonomy, no.
- No seven-figure SaaS for our portfolio. The lab does not extract rent from the cohort.
- No vanity ventures. If we can't see a path to a working business, we kill the engagement before we co-found.
Research partners
We co-publish and run joint research with universities and operators across Europe, the Middle East, and North America. Some partnerships are formal multi-year framework agreements; others are informal — co-authored papers, sabbatical exchanges, shared datasets.
- Texas State UniversitySoilless agriculture in stressed-water environments · anchor partner for Babylonian since 2022
- Boğaziçi UniversityAcoustic ML & signal processing · joint research since 2023
- Universidade de LisboaMarine robotics & ocean engineering · framework since 2024
- USPTO open dataPublic reaction corpus underpinning ChemSynth
- Operator partnersNorth-Sea wind operator, Norwegian salmon farms, kitchen-equipment OEMs, NATO-aligned integrators