AVA: AI Copilot & Autonomous Delivery Pipeline
An internal AI copilot and autonomous delivery pipeline used across multiple DubizzleLabs products, letting business and product teams query live data and ship code end to end from a single chat interface.
PR prep time: ~40 min → ~6 min
The problem
Product and business teams at DubizzleLabs could not answer their own data questions, and every feature request had to be translated by an engineer before any work started. That translation step (question to report, idea to ticket, ticket to branch) consumed senior engineering time that had nothing to do with engineering.
How it works
- A single sign-on chat interface spanning products, backed by product-specific MCP servers pulling live data from databases and OpenSearch to generate custom reports on demand.
- A requirements flow: a team describes a feature in chat, AVA asks clarifying questions, drafts wireframes from the internal knowledge base and an LLM, and files a Jira ticket with the wireframe attached.
- An autonomous delivery pipeline chained onto that ticket. A Claude skill turns it into an engineering plan, opens a branch and GitHub issue, then triggers a workflow that generates code and tests, opens a PR, comments back on Jira, and notifies QA.
- QA closes the loop, verifying the generated changes with their own Claude skill before merge.
What changed
PR preparation on affected workflows dropped from an average of ~40 minutes to ~6 minutes, and non-engineering teams got a direct, self-serve path from a data question or a feature request to shipped code. The constraint that mattered was trust: every generated change still passes through human review and QA verification, which is what made the teams willing to adopt it.