How we work
From a loose idea to solid software.
Four stages, with working versions you can click along the way. No black boxes, and no big reveal at the end.
Four steps, start to finish.
From a loose idea to solid software in four steps — and you can click it at every stage.
Discover
Liquid — still taking shapeWe learn how your business runs and find where AI creates real value — and where it doesn't.
You get a scope, a roadmap and an estimate- What happens
- Calls with the people who do the work today. We look at your data, your tools and the questions that repeat.
- Who you talk to
- The engineers who would build it.
Prototype
Gel — it holds a formBefore the big build you see it working: a clickable prototype, or an AI demo running on your own data.
You get something you can try- What happens
- The riskiest part is built first, on your real documents or data, so everyone can see whether the idea holds.
- Who you talk to
- The engineers, plus the people on your side who will use it.
Build
Solid — engineered and testedShort cycles and reviewed code, with working versions on a staging link as it takes shape. No black boxes.
You get production-ready software- What happens
- Planned architecture, code written and reviewed by engineers, tests and evaluation sets that run on every change.
- Who you talk to
- The engineers building it, with progress on a link you can click.
Launch & scale
Crystal — it keeps growingWe deploy, monitor and keep improving features, models and prompts as your users grow.
You get support after go-live- What happens
- Deployment, monitoring and training for your team, with documentation and ongoing support.
- Who you talk to
- The engineers who built it, for fixes, improvements and model updates.
How long will it take? It depends on scope. After a short discovery call you get a realistic timeline and an estimate — before anything is built.
Start a projectProgress you can click.
Instead of status reports you get a staging link. Here is what a few versions of a document-automation build look like.
- v0.1Demo 1Upload one of your own invoices and see what the model pulls out.
- v0.2Demo 2Sign-in, roles, and a list of every document received.
- v0.3Demo 3Extraction on your real documents, shown on a review screen.
- v0.4Demo 4Checks against your purchase orders. Mismatches wait for a person.
- v1.0Go-liveDeployed and monitored, with documentation and support.
AI-native engineering
AI-native. Engineer-approved.
Our engineers work with AI tools every day — to draft, test and document faster. You get the speed; nothing ships until an engineer has reviewed it, the tests pass, and you've tried it yourself.
- AI drafted the upload screen and its APIFrom the engineer's brief · 4 minutes
- RKReviewed line by line — 14 changes, edge cases addedEngineer · security and error handling checked
- 128 automated tests passedRuns on every change
- Ready for you to trystaging.your-project.app
Illustrative · fictional feature
Launch isn't the end.
After go-live we keep watching, fixing and improving — and your team gets the training and documentation to run it. Here's what that looks like a few months in.
Recent updates
- v1.4Faster search on large catalogues2 days ago
- AIModel updated — answers noticeably quickerLast week
- SecurityLibraries patched, nothing for you to doLast week
- v1.3New monthly report for the finance team3 weeks ago
Your team
Illustrative · fictional app and numbers
Timelines and pricing, honestly.
Every project is different, so we scope it before we quote. After a short discovery call you get a realistic timeline and an estimate — and a small prototype is often the cheapest way to find out what's worth building.
Let's build what's next.
Tell us what you need in a few lines. An engineer reads every brief and replies with questions, a plan and an estimate.
- Replies from engineers
- NDA on request
- Clients worldwide