Technology
Calculate where possible. Interpret where needed. Validate throughout.
Our technical approach favours well-understood tools, explicit data flows and evidence over novelty.
Engineering principles
Decisions that stay reviewable.
- 01
Deterministic core
Business rules, calculations and transformations live in explicit, testable code.
- 02
Bounded AI
Models handle interpretation tasks within measured limits and never silently overwrite facts.
- 03
Typed boundaries
Schemas and contracts at every interface, validated on the server.
- 04
Observable by default
Logs, metrics and traces designed with redaction and retention in mind.
Stack
Chosen per project, from a small set we know deeply.
We prefer mainstream, long-lived technology so that clients can hire for it and maintain it.
Application platforms
TypeScript with React and Next.js, Python services, .NET where the environment calls for it.
Data
PostgreSQL first; object storage, search and streaming systems when the workload requires them.
AI and machine learning
Hosted and self-managed language models, classical ML, retrieval pipelines and evaluation tooling selected per data-protection requirement.
Infrastructure
Containers, infrastructure as code, European cloud, dedicated hosting and a Caddy-based reverse proxy.
Quality
Evidence travels with the software.
Automated tests, static analysis, dependency audits and reproducible builds are part of every delivery. Security headers, input validation and rate limiting are treated as baseline, not as add-ons.
Public system view
A modular path from input to reviewable output.
- Input from a person, a document or a connected system
- Deterministic processing and validation against explicit rules
- Optional AI assistance for bounded interpretation tasks
- Human review states where consequence requires it
- Structured outputs, integrations and audit records
Discuss an architecture before it becomes expensive.
Early technical conversations are the cheapest place to remove risk.