We build software for organisations that need it to work, and products of our own. AI runs through both.
We take a narrow set of work and go deep on it, rather than claiming everything.
Models put to work inside real systems — retrieval over your own data, decision support, and automation of the work that actually consumes people's time. Built to be measured rather than demonstrated.
Backends, data pipelines, and interfaces that run unattended. Instrumented and observable, and written so the team that inherits them can maintain them.
From a first version to something people depend on daily. We do this for clients and for ourselves, which changes how we scope and what we argue against.
Getting systems to talk to each other — telemetry, third-party services, legacy platforms — and turning what comes back into something a person can act on.
Measured, not assertedAgriculture at field scale, cross-border tax, and regulated financial infrastructure.
A digital farming platform. In-field IoT sensors, dendrometers measuring plant stress directly, and satellite imagery feed AI irrigation recommendations used by over a thousand growers worldwide, with automated irrigation control.
Automated VAT and GST reclaim across more than forty countries, with AI that validates accounts-payable invoices against each country's tax rules before payment rather than after.
Open Finance — Israel's regulated financial-data sharing regime, the local counterpart to PSD2 open banking, giving licensed providers consented access to customer financial data under central bank supervision.
Running our own product changes how we build for everyone else — we carry the consequences of our own decisions.
A management platform for gyms, studios, and personal trainers: scheduling and bookings, member management, memberships and billing, a sales pipeline, and a member-facing mobile app. Built, run, and supported by Billie-Tech.
kalistera.com →These are stances rather than promises about timelines — the shape of an engagement is set with you before it starts.
We would rather deliver a smaller thing that works than a larger thing that nearly does. The first conversation is usually about what to cut.
Real data, real load, real failure modes, from early on. A demo that only works on the happy path tells nobody anything useful.
Documented, instrumented, and written to be read. You should not need us in order to keep running what we built.