Engineering
Engineering standards and technical evidence
We show how we work without publishing confidential source code: the principles we follow, the checks that enforce them, and the infrastructure we use.
Standards
- 01
Reproducible by design
The same input produces the same result. Renders, builds and releases are traceable and repeatable; time and randomness are controlled explicitly.
- 02
No deploy without tests and gates
Automated tests and quality gates run before anything reaches production, and a failing check blocks the release.
- 03
Secrets stay out of code
Credentials live in managed environment configuration, never in repositories or logs.
- 04
AI narrates, systems compute
Distances, areas, prices and dates come from deterministic code; language models only phrase them.
- 05
Provider independence
Imagery, AI and infrastructure providers sit behind our own interfaces and can be replaced without a rewrite.
- 06
Documented handover
Architecture notes, decision records and runbooks are delivered with the code.
Test and quality approach
- Unit, integration and end-to-end tests run before every deployment.
- Negative controls: each quality gate is tested against the defect it is meant to catch.
- Baseline and equivalence tests when changing a working pipeline, so behaviour changes are deliberate and reported.
- Measured claims only: performance and quality statements come with the command or data that produced them.
Operating what we ship
- Health endpoints that separate errors from warnings, so alerts mean something.
- Alerts to the team by e-mail and Telegram.
- Every release is measured after deployment — status, health, connected workers — before it is called done.
- Backups before data migrations, and a rollback path for every infrastructure change.
Infrastructure and tools we use
| Area | Tools |
|---|---|
| Web | React, Next.js, Astro |
| Mobile and desktop | Flutter, Tauri, Swift |
| Backend | Node.js (NestJS), Python (FastAPI) |
| Data | PostgreSQL, SQLite, Prisma, Cloudflare D1 |
| Geospatial | Mapbox GL, elevation models, cadastral and open map data |
| Rendering and media | Playwright, ffmpeg |
| AI | OpenAI, Anthropic, Google Gemini, local models via Ollama |
| Infrastructure | Cloudflare (Workers, R2, Tunnel), Coolify on VPS, GitHub |
Evidence packs
For performance and quality claims we prepare an evidence pack: the measurement, the raw data, integrity hashes and a script that recomputes the result. The LLM Studio benchmark is the first; others are shared under NDA.
Have a project in mind?
Send a short description. If you need a mutual NDA first, we will send ours before you share details.