An LLM brain with real tools
The model reasons; durable tools fetch data, write code, run checks, open pull requests, and report evidence.
01 / Shipped work
No anonymised logos and no padding: four cards, each naming the system, linking to its architecture and release path, and saying plainly whether it is a client delivery or a product we build and run ourselves.
IndepAI
Deterministic FIRE math the model is never allowed to compute, Supabase-backed accounts, and a coach that reads only what the user has consented to share.
HSK WebKD
Access-control and time-tracking devices behind NestJS modules, Postgres and Redis. A Helm migration job gates readiness, so a schema change cannot reach production half-applied.
Hanbud
Four separate buyer paths instead of one generic contact form: catalog discovery, product detail, cart, and B2B offer request, in Polish and Ukrainian.
Kontenero
A quote flow that takes address, waste type, volume and callback time, then routes it only to operators with an active BDO permit. Our build, and not finished yet.
02 / Agent infrastructure
We build agents the way we build production services: tools, memory, review gates, telemetry, rollback. Your team runs it after we leave. It isn’t a demo that dies after the pilot.
Skills encoded from your company
Your sales, support, operations, and engineering patterns become reusable agent skills with clear boundaries.
Human gates stay explicit
Agents inspect, draft, test, summarize. Anything touching production keeps approvals, logs, and a rollback plan.
03 / How we work
Built for companies with devices, data, legacy systems, compliance pressure, and AI ambitions. Teams that still have to ship every week.
Model the domain
Actors, data, jobs, devices, queues, approvals, owners, all named before the first line of code.
Build the slice
One route, one workflow, one integration, one release path. The smallest real system beats the broadest slide deck.
Gate autonomy
Agents accelerate discovery, writing, testing, documentation. Review boundaries stay explicit and human.
Leave proof behind
Every engagement ends with diagrams, runbooks, dashboards, and prompts your team keeps.
Free · written process review
Describe the process in four fields. Within one business day you get a written process review: what we would automate first, which systems it touches, and the first test that would prove it works. No call required to get it.
If the review points to a build, the next step is the AI workflow pilot: a fixed-price, two-week build, quoted before kickoff.
04 / Focus areas
The sweet spot: product thinking, backend architecture, cloud operations, and AI agent workflows meeting in one hard problem.
AI products with deterministic business logic
Autonomous agents for company operations
Kubernetes and Helm production platforms
NestJS backends and integration layers
Design systems and conversion pages
Documentation ready for a public case study
06 / Next step
Start with the free written process review. If it points to a build, that build is the AI workflow pilot: a fixed-price, two-week build in which we map the domain, ship one process end to end, and leave the documentation with you whatever you decide next.
- Written review back within one business day
- No sales sequence
- Fixed price for the pilot, quoted before kickoff
fixed price, quoted before kickoff · two weeks · documentation stays with you
Tell us what is slow
Four fields. You get a written process review and what we would automate first.
