Summary

• Multi-provider transcription and summarization (OpenAI + Gemini) with provider selector UI
• AI summarization via Chat Completions API, stored as sidecar markdown files with tab toggle
• Obsidian export via URI scheme with remembered vault config
• Duration limit — transcribe only the first N minutes
• Chunk cache resume — interrupted transcriptions skip already-completed chunks
• ETA progress with per-chunk timing in SSE events
• Gemini retry logic (3 attempts, exponential backoff)
• SVG liquid wave animation replacing CSS bars
• Integration test overhaul with LLM judge quality validation

Test plan

• poetry run pytest -v — all unit + endpoint tests pass
• poetry run pytest -m integration -v — integration tests pass with real API keys
• Transcribe multi-chunk video, cancel mid-way, re-transcribe — verify chunk resume
• Switch providers in UI — verify correct models used
• Test Obsidian export, duration limit, summarization

🤖 Generated with Claude Code