DRANSAY · AI
An AI Hub for Product and Engineering at DrAnsay
- Company
- DrAnsay
- Period
- Nov 2025 to present
The AI Hub is a set of role plugins, repository context and a company knowledge search, built to bring AI tools into everyday delivery in Product and Engineering at DrAnsay. The design gives each role workflows that produce the documents it is accountable for, has each repository describe its own context, and lets AI tools reach company knowledge through one entry point. A named person reviews and approves what the tools produce.
Today Product Managers, team leads and developers use the plugins, the repositories’ dependency manifests feed one company-wide graph, and other departments have adopted the knowledge search. What comes next is written down as a roadmap.
Role-based delivery plugins
The centre of the hub is one repository of delivery plugins: a shared Core plugin, four role plugins for the Product Manager, Engineering Lead, Engineer and Quality Engineer, and a maintainer plugin for changing the catalogue itself. The same workflows run in Claude Code, Codex and Cursor. Most people need only Core and their own role’s plugin.
The plugins follow one operating model. Work moves from an idea to a product requirements document, a delivery plan, engineering tasks, behaviour specifications, implementation, a verification report and a pull request. Each document has one accountable role. The Product Manager owns the requirements, the Engineering Lead owns the plan and any design decision records, the Engineer owns the behaviour specification and the pull request, and the Quality Engineer owns the test plan and the verification report. Design records are optional: the Engineering Lead writes one only when a decision crosses team boundaries or needs detailed design, and records why when it is skipped. A model may draft or review any of them; accountability stays with the person. Writing files and changing external systems such as the ticket tracker both wait for an explicit approval.
The design draws on public specification-driven tools such as spec-kit and OpenSpec, adapted to how DrAnsay plans and delivers. I wrote most of the plugins myself, and two colleagues contributed.
Context per repository
Each repository carries its own context in files committed beside its code, so every contributor and every agent starts from a configured repository instead of a personal setup.
A second file, the dependency manifest, declares what the repository depends on and what depends on it: the services it calls, the events it publishes and consumes, with the evidence for each relationship and how confident it is. The design asks the owning team to update it in the same pull request that changes the code. I designed the manifest format and, in November 2026, rolled it out across the company’s repositories, seeded from the existing architecture catalogue and checked so that the combined graph stayed unchanged.
Aggregated graph
The manifests roll up into one company-wide dependency graph. Where a repository has no manifest yet, a central catalogue still describes it. The living architecture map is a view on that graph: an interactive, animated map of the apps people open, the backend services, the event bus, the data stores and the third-party platforms, with the messages that travel between them. Guided stories replay real flows step by step. I built the map.
DrAnsay MCP
The DrAnsay MCP is the entry point through which AI tools reach company context. It does two jobs. It indexes company knowledge, and it routes a request, such as a ticket key or a link, to the tickets, specifications and documents that belong to it.
Company knowledge search
The company knowledge search puts the company’s written knowledge behind one search, for people and for AI tools. The AI Engineering team set it up, and other departments later adopted it. A report-only audit automation, built in September 2026, reviews the content so the index does not amplify stale or duplicated pages.
Beside it, a documents portal makes the company’s specifications and decisions searchable in one place, filtered by type, team and initiative, while each document stays in its original source. I built the portal.
Roadmap and AI Guild
I wrote the roadmap for what comes next: work items grouped by month, each scored for effort and impact. The themes are company knowledge, a shared environment for automations, quality and reliability automations, and the evaluation of agents.
The roadmap sets one safety rule for every automation on it: one that writes to internal systems needs a person to approve each write until its evaluations pass, and one that touches production or external systems needs a security review first.
An AI Guild is planned to own the roadmap.