Private MVP for semantic QA intelligence using local-first browser architecture.

• Upload Azure DevOps-style test case data (.json or .csv)
• Generate synthetic 10,000 test cases for MVP benchmarking
• Compute KPI dashboard in-browser
• Build embeddings and semantic clusters
• Ask QA Copilot questions over clustered context

1. Upload Layer
2. Vector Layer (local store, Zvec adapter-ready interface)
3. Analytics Layer (DuckDB-WASM)
4. Reasoning Layer (Qwen/WebLLM primary, template fallback)

• Redundancy Score
• Entropy Score
• Orphan Tag Ratio
• Exact duplicate groups
• Near-duplicate groups

• Semantic cluster scatter map
• Suite distribution chart (DuckDB query backed)

• test_case_id
• test_plan_id
• test_suite_id
• title
• description
• steps
• tags

npm install
npm run dev
npm run build
npm run test:quality

• Unit + evaluation tests (Vitest)
• End-user live UX tests (Playwright)
• Automated output files stored under results/:
  • results/TEST_RESULTS.md
  • results/TEST_RESULTS.csv
  • results/eval_results.csv
• LLM evaluation artifacts:
  • eval_strategy/expected_results.md
  • eval_strategy/actual_results.md

• Embedding model attempts Xenova/all-MiniLM-L6-v2; if unavailable, falls back to deterministic hash embeddings.
• LLM engine attempts WebLLM Qwen runtime; if unavailable, uses deterministic reasoning fallback.
• Everything runs fully in browser (no backend, no API keys).

• Replace local vector implementation with direct Zvec browser runtime binding.
• Add coverage heatmap and semantic family drill-down panel.
• Add execution data phase (flakiness and runtime intelligence).