Build the models that stop ATM card skimming.
2.1M terminals, 70+ markets. We read the machine, score the withdrawal inside the authorization path, and hand the evidence to the people who act on it.
- Terminals monitored
- 2.1M
- Markets live
- 70+
- Roles open
- 4
- P50 decision
- <40 ms
A small team on a physical problem.
Remote-first and async by default. Everyone spends time with the hardware, because the models run against terminals in the street and not against a spreadsheet.
Remote-first, globally
Work from wherever you do your best work. We are async-friendly and outcome-driven.
Meaningful equity
Every team member shares in the upside of what we are building together.
Premium health cover
Comprehensive medical, dental, and vision for you and your dependents.
Learning budget
An annual stipend for courses, books, and conferences to keep growing.
Real time off
Generous, genuinely encouraged paid leave — plus company-wide recharge days.
Work that is physical
Our models run against hardware in the real world. Field weeks are part of the job, and the team goes together.
4 roles open right now.
Each one owns a piece of the detection path. If none of them fit, send us the work you have already shipped and tell us what you would build next.
Own the models that decide whether a machine is compromised. You will work on the boundary between device telemetry, transaction streams, and the adversary who is trying to look normal.
What you own
- Train and ship sequence models over authorization streams for skimming and cash-out signatures.
- Build evaluation sets from confirmed skimming incidents, not just labelled fraud.
- Take models from notebook to a sub-40 ms decision path and keep them there.
- Work with analysts to turn every closed case into training signal.
What you bring
- 3+ years shipping ML to production with measurable impact.
- Strong Python and SQL; comfortable with streaming, imbalanced and adversarial data.
- Experience with anomaly detection, sequence models, or fraud systems.
- Bias toward evaluation rigour over leaderboard scores.
Teach our models to see hardware that should not be there. Overlay skimmers, inserts, false keypads, and cash traps — across thousands of terminal models, lighting conditions, and camera generations.
What you own
- Build detection models for reader-surface and fascia anomalies on real terminal imagery.
- Handle domain shift across vendors, markets, and low-quality camera feeds.
- Own the labelling pipeline and the on-device versus cloud trade-off.
- Ship visual evidence that a human analyst can verify in seconds.
What you bring
- 3+ years in CV or perception with production deployments.
- Strong PyTorch and a working knowledge of detection and segmentation architectures.
- Experience with edge inference or severe latency budgets.
- Comfortable building datasets from scratch.
Keep a decision API inside the authorization path of banks in 70+ markets. Latency is the product; availability is the contract.
What you own
- Operate the scoring path at single-digit-millisecond overhead across regions.
- Own observability, on-call, and the incident process for tier-one customers.
- Harden the platform to SOC 2 and data-residency requirements.
- Automate deploys, capacity planning, and failure drills.
What you bring
- 5+ years in SRE, platform, or infrastructure engineering.
- Deep experience with cloud infrastructure, containers, and IaC.
- Track record running low-latency services with strict availability targets.
- Security-minded; comfortable owning compliance-relevant systems.
Put detection into production at issuers, networks, and fleet operators. You will run the integration that turns a proof of concept into live scoring.
What you own
- Run technical discovery and integration with fraud, switch, and operations teams.
- Translate a fleet's incident history into a detection plan with measurable targets.
- Own the path from sandbox key to production decisioning.
- Feed field findings back into the product and the detection models.
What you bring
- 3+ years in solutions engineering or technical presales, ideally in payments or fraud.
- Comfortable with APIs, streaming data, and reading a bank's architecture diagram.
- Excellent written communication across time zones.
- Willing to travel to terminals when the problem is physical.
Open application: tell us the problem you want to own at ATMList, and what you would do in your first month.
Get in touch