Internal knowledge, inside the IDE

A SaaS B2B business working with Cloud Combinator on AWS. The client is anonymised at their request.

The challenge had four focus areas.

The engagement was delivered as an infrastructure-as-code proof of concept, built to a secure development standard in the client's own AWS environment and structured around clear milestones.

Acceptance is defined against a working, maintainable system: the API queries the knowledge base and returns answers, ingestion syncs automatically, the MCP server lets a coding agent enrich its context on demand, and the client's team can run it themselves. The items below reflect the deliverables and design set out in the Statement of Work.

• Knowledge where developers work — Coding agents can pull grounded answers from the client's internal knowledge base on demand, without leaving the IDE.
• Always current — Event-driven ingestion keeps the knowledge base in step with its Amazon S3 source automatically, so answers reflect the latest documents.
• Guardrailed and grounded — Amazon Bedrock guardrails and prompt engineering keep responses safe and relevant, drawing on the client's own content.
• Traceable by design — Invocations are logged to DynamoDB, giving the client's auditability over what the assistant is asked and what it returns.
• Handover — The proof of concept closes with a runbook, a reference architecture, a knowledge-transfer session and sign-off that the client can maintain and iterate on the system themselves.

With internal knowledge now reachable from inside their coding agents and a foundation they can operate themselves, the client leaves the proof of concept with a repeatable pattern for AI-assisted development, ready to broaden the knowledge base and roll the experience out more widely across their engineering team.