AI & automation

One engineering brain for a whole company

A private AI layer over every repo, ticket, runbook and dashboard, with agents that act under approval.

Client: Product companies, 40 to 400 engineers
Field: AI & automation
Our part: Full cycle: audit, architecture, rollout, hand-over
Stack: PythonTypeScriptpgvectorKubernetesClaude and open-weight modelsTemporalOpenTelemetry

The problem

Knowledge about why a system behaves the way it does lives in old tickets, chat threads and two people’s heads. Every incident, onboarding and estimate pays for that again. Chat assistants bolted on the side do not help: they do not know the codebase and cannot do anything with the answer.

What we built

A private index across repositories, issue trackers, runbooks, metrics and post-mortems, permission-aware down to the file
An answer engine that cites the commit, ticket or dashboard it drew from, so an engineer can check it in one click
Agents for triage, on-call summaries, release notes and routine ops queues, each with an approval step and an audit trail
Evals that run on every prompt and model change, so quality is a number, not a feeling
Model routing and cost control across hosted and self-hosted models, deployed inside the client’s cloud

What it changed

Engineers ask the brain first and get an answer with sources instead of paging the one person who remembers
Ops teams hand routine queues to agents and approve the exceptions
The client owns the data, the prompts and the evals; we hand over the keys with runbooks

Next case

Slot mechanics for a licensed casino