AI-Powered Testcase Generator (TestCraft AI)
QA engineers and product teams often spend significant time converting requirement documents into manual test cases while maintaining traceability to source specifications and accounting for previous defects. Because requirements, bug reports, and existing test cases are frequently stored in separate systems, important regression risks can be overlooked when new test coverage is created.
To address this challenge, I built TestCraft AI, a tool that generates structured, review-ready manual test cases from requirement documents. Users can upload requirements and optionally import historical bugs and existing test cases. Documents are prepared through chunking and embedding, then processed by a LangGraph-based multi-agent workflow. The pipeline performs requirement analysis, scope-aware retrieval, parallel test generation, optional coverage review, deduplication, and persistence. Retrieval-augmented generation (RAG) using Supabase pgvector surfaces relevant historical context before test cases are generated, helping the system produce coverage that reflects real regression risks rather than generic templates. Streamlit provides the user interface for ingestion, generation, semantic search, traceability reporting, and CSV/XLSX export, while LangSmith tracing can be used to inspect and evaluate pipeline runs.
A key learning was that scope-aware retrieval made generated test cases more relevant by using bugs and existing tests from the same UI/service scope instead of the entire document. LangGraph was a great fit for orchestrating analysis, RAG retrieval, generation, and coverage checks in one workflow. In future iterations, I would add persisted atomic requirements, retrieval reranking, and human-in-the-loop checkpointing within the LangGraph workflow.
