QualGent AI Assistant
An enterprise QA copilot routing across 45+ tools and sub-agents from a single orchestrator.
- Software Engineer Intern
- QualGent (YC X25)
- 2025
- Shipped
45+
Problem
QA teams had capability scattered across a dozen surfaces: test history in one place, tickets in another, app knowledge in a third. Every question meant stitching them together by hand.
Approach
I built QualGent's flagship assistant: one orchestrator that could reach everything. The interesting engineering was in routing, not generation. With 45+ tools available, picking the right three is the whole problem.
What I built
An enterprise QA copilot built as a Gemini 2.5 Pro orchestrator on Google ADK, routing across 45+ tools and sub-agents and deployed on Vertex AI Agent Engine through a gated Cloud Build pipeline.
Architecture
Gemini 2.5 Pro orchestrator on Google ADK, routing across 45+ tools and sub-agents with shared state management.
Tool surface spanning RAG retrieval, an MCP Postgres/pgvector Toolbox, and Jira and Linear through OAuth.
Deployed to Vertex AI Agent Engine via gated Cloud Build, so agent changes ship through the same review gate as code.
Impact
Unified 45+ autonomous agents and tools behind one interface, becoming the primary way QA engineers at QualGent queried app knowledge, test history, and issue trackers.
45+