QA Agent: AI test-generation and execution platform
- Type
- Project
The QA Agent reads a Jira ticket or pull request, generates test cases, and runs them in gVisor-sandboxed, network-policy-isolated Kubernetes Jobs, then judges pass or fail with an LLM and reports through a React and TypeScript dashboard I built.
It runs a genuine ReAct-style LangGraph loop rather than a fixed pipeline, with test generation grounded by retrieval against an internal context service and orchestration handled through Temporal for durability. Ollama runs local models as a first-class provider alongside direct Anthropic API integration, keeping inference cost down without sacrificing coverage.
The whole system runs inside an ISO 27001 certified environment with live GCP and GKE access, backed by a hand-rolled Postgres migration layer and verified with a full pytest and Vitest suite.
Built with
- LangGraph
- Retrieval-Augmented Generation (RAG)
- Temporal (durable workflows)
- gVisor sandboxing
- Kubernetes
- GCP / GKE
- Neo4j
- Docker
- ISO 27001 environment
- Python
- FastAPI
- PostgreSQL
- React 19
- TypeScript
- Vite
- TanStack Query
- React Router
- Vitest
- Mock Service Worker (MSW)
- oxlint
- pytest
- Ollama
- Qwen Coder
- DeepSeek Coder
- Anthropic Claude API (direct SDK)
Related