Introducing TalentSuite Atlas: documentation that writes itself from your code
Every digital service team knows the feeling. The code moves on every sprint, but the documentation describing it was last touched months ago. Ask how two services talk to each other and the answer is usually a person, not a page.
TalentSuite Atlas is our answer to that problem. It reads your code and generates the documentation for you, then keeps it current every time the code changes.
The problem: knowledge that lives in people's heads
Large digital services rarely sit in one codebase. They span dozens of interconnected .NET repositories: core APIs, back-end processing, data services and user-facing portals. How those pieces fit together is often understood by a handful of experienced people and written down somewhere that has drifted out of date.
That gap has real costs:
Slow onboarding. New joiners spend their first weeks reverse-engineering what already exists.
Risky cross-service change. Knowing who consumes an event or calls an API depends on whoever remembers.
Hidden legacy risk. Finding which services run on .NET versions nearing end of support means checking repositories one by one.
Harder assurance. Data protection reviews, security audits and service assessments all start by asking where things are and how they connect.
How TalentSuite Atlas works
Atlas is a set of AI agents hosted on Azure AI Foundry. They work through your repositories in four steps:
Manifest. A configuration repository lists each repository to scan and where to look for each type of artefact.
Discover. A discovery agent selects the relevant source files from each repository.
Generate. A generation agent, powered by Azure OpenAI, turns those files into structured documentation.
Review. Everything lands as a single pull request on the catalogue repository, ready for a person to approve.
Atlas is read-only on your code. It never changes the repositories it scans, and everything it produces goes to its own catalogue. Only repositories that have changed since the last run are processed again, tracked by commit hash, so regular runs stay quick and cheap.
Bringing a new service into scope is a configuration change, not a development task. Add the repository to the manifest and the next run picks it up.
What it generates
Atlas produces seven types of artefact, all read directly from the source code so nobody has to write them up by hand.
Artefact | What you get |
API contracts | OpenAPI 3.1 specifications for each API controller |
Data models | Tables, relationships and indexes, with likely personal data flagged |
Event catalogue | Messages, their fields, and the handlers that consume them |
Dependencies | What each service calls, with no secret values extracted |
C4 diagrams | Context and container views as editable draw.io files |
Local dev setup | The services and configuration needed to run each repository locally |
Repository metadata | .NET versions mapped to Microsoft's end-of-support dates |
Repository metadata is fully deterministic. It parses project files directly, with no AI model involved.
Where it pays back
Onboarding onto a team. A developer joins a team that owns a core service. Instead of relying on colleagues for walkthroughs, they open the catalogue entry and find the API specs, database schema, events, dependency map and local setup notes. Their early questions are about the work, not about what exists.
A change that crosses services. When a change to a business process touches several services, the event catalogue shows every service that publishes or handles the affected events. The impact list is ready before refinement, which means better-scoped stories and fewer surprises in testing.
Getting ahead of legacy .NET. Atlas maps every target framework to its end-of-support date across the whole estate at once. The result is an upgrade backlog ordered by support date, and hard evidence for roadmap and funding conversations.
Data protection and assurance. DPIAs, security reviews and audits start from a map of where personal data likely sits and which external systems each service talks to, rather than a blank page.
Mobilisation and handover. Taking on an unfamiliar estate, or rotating teams, needs a credible picture fast. Add the repositories to the manifest, run the pipelines, and architecture, contracts and the risk picture arrive together in one reviewed pull request.
External assessment. When an assessor asks where the documentation is, including for services nobody actively owns, you can point them at a catalogue generated from the code itself.
Built with safeguards, and honest about its limits
We designed Atlas so that nothing is published without a person checking it:
Every change lands as a pull request for human review.
Secret values are never extracted or stored.
Atlas only reads your repositories and never writes to them.
We are also upfront about what it cannot do. Generated artefacts can be wrong and need spot-checking. Personal data flags are a best-effort judgement to support a reviewer, never a compliance sign-off. Output quality depends on an accurate manifest.
Running costs are low. Atlas uses pay-as-you-go Azure OpenAI, and our indicative estimates put nightly runs across 100 repositories at roughly £35 to £210 a month, not thousands. Running weekly instead of nightly brings that down further.
A living map of your service estate
TalentSuite Atlas gives teams documentation that is generated from the code, reviewed by people, and kept current for pounds a month. It works for any GitHub-hosted .NET estate, whatever the domain.
If your teams are losing time to out-of-date documentation, or you are preparing for a handover or assessment, we would love to show you Atlas in action. Get in touch with the Talent Consulting team to arrange a demo.
