v1.4.0 - Frontend Improvements

✅ Keyboard Shortcuts

• Ctrl/Cmd + K: Focus query input
• Ctrl/Cmd + Shift + K: Toggle dark/light theme
• Ctrl/Cmd + Enter: Submit query
• Esc: Close dialogs

✅ UI/UX Enhancements

• Resizable query textarea (min-h-12 with resize-vertical)
• Search functionality in chunk viewer
• Copy-to-clipboard buttons with visual feedback
• Better loading states and success indicators
• Improved visual hierarchy and spacing

✅ Better Error Handling

• Error boundary component
• Retry logic in API calls
• Enhanced error messages with toast notifications

✅ Component Improvements

• Document list with keyboard shortcuts info
• Query panel with tabs for response/retrieval/prompt analysis
• Upload panel with success/error visual states
• Skeleton loaders for better UX

v1.3.0 - Testing, guardrails, and output grounding

• Added pytest suite: unit tests for chunking, extraction, guardrails, and full API endpoint tests (LLM calls mocked, tests run free and deterministic)
• Added retrieval evals: a golden question set that measures whether semantic search actually surfaces the right information, separate from unit correctness tests
• Added request guardrails: file type/size limits, query length limits, top_k bounded to 1-20
• Added an output grounding check: flags when the model's answer doesn't actually overlap with the retrieved context, surfaced as a badge in the UI
• Fixed a real bug caught by the new test suite: overlap=0 was silently falling back to the default overlap due to Python falsy-value handling
• CI now runs pytest and the eval script on every PR, not just import checks

v1.2.0 - PDF and DOCX support

• Ingestion now supports PDF and DOCX files, in addition to txt/md
• Clear error handling for unsupported file types and unreadable/empty documents
• Frontend surfaces the actual backend error instead of a generic failure message

v1.1.0 - Documentation and cleanup

• Added project README covering architecture, setup, and API reference
• Documented branching strategy and disaster-recovery process
• Removed leftover debug logging

v1.0.0 - Core RAG pipeline

Initial working release.

• FastAPI backend with document chunking (configurable size/overlap)
• Chroma vector store for local, persistent embeddings
• OpenAI-compatible LLM client for answer generation
• Clear-on-ingest and clear-on-demand controls
• Next.js inspector UI: upload, chunk viewer, query with retrieval/prompt/response tabs
• CI pipeline validating backend imports on every PR