AI in products and in software delivery
We build language models into products where they add real value, and work with AI agents every day — so code is better tested and documented, not just written faster.
- 01LLM integrations with multi-provider fallback
- 02Assistants and actions built on your system's API
- 03AI agents in legacy-code modernisation
- 04Tests, documentation and code knowledge graphs
We treat AI like any other technology: first the question of which problem it solves, then cost, risk and a plan for when it fails.
AI in products
- Language-model integrations — translation, classification, data extraction, summaries. With several providers and automatic failover when one does not respond.
- Assistants for your data — an OpenAPI specification and an intermediate layer that let an AI assistant use your system's API safely and within limits.
- Process automation — models as a step in a business process, with a human approving wherever a mistake is costly.
AI in software delivery
We work with AI agents when modernising legacy systems, writing tests and documentation. Our experience differs from the popular view that AI lowers code quality: in the hands of an experienced architect it makes possible work that used to be too laborious. In one modernised service, the test count grew from zero to over two hundred, and the documentation from nothing to several dozen documents. We describe this in detail in Insights.
Related reading
2026-10-10
AI does not ruin code. In an architect's hands it raises quality
The market consensus says language models flood repositories with lower-quality code. Our experience modernising a large platform is the opposite — on one condition.
2026-10-18
BiLSTM with attention on market data. The hard part is an honest experiment, not the network
Over several months we built BiLSTM networks with attention, a reinforcement-learning environment and a point-in-time fundamentals layer. What paid off most were not the architectures but placebos, thresholds fixed in advance and written-up negative results.
Wondering where AI could actually help?
Point us to a process or the data you work with. We will check whether a language model adds measurable value there — and what it will cost to run.