Can I vibecode Sembly AI?

text
Build a comprehensive personal replacement for Sembly AI in an empty repository, focusing on generating structured minutes and deep speech understanding.
Use Python 3.12, FastAPI, SQLite, whisper.cpp, and a clean HTMX + Tailwind CSS interface; do not offer alternative stacks.
The core job to be done: Record or import a meeting audio file, transcribe it locally with high accuracy speaker diarization, allow manual speaker correction, and automatically produce structured meeting minutes, actionable tasks, and a list of discussed topics.
Make the first run work locally with one documented command. Store all user data securely and locally in SQLite by default.
Put API keys for any optional LLMs (used for advanced summarization and topic extraction) in .env, ship .env.example, and never commit credentials.
Add explicit start, pause, resume, and stop controls with a visible recording indicator in the web interface.
Support importing WAV, MP3, M4A, and MP4 files, utilizing ffmpeg for audio extraction.
Run transcription locally and display the output as timestamped, editable segments. Let the user rename speakers globally (e.g., "Speaker 1" -> "Alice") and propagate those corrections throughout the entire transcript.
Implement an "Insights Engine" that passes the transcript to a local or API LLM to generate: a high-level executive summary, key decisions made, a bulleted list of action items, and overarching topics/themes discussed in the meeting.
Export capabilities are crucial: allow exporting the structured notes and transcript to Markdown, plain text, and WebVTT beside the original recording.
Include clear empty, loading, success, and recoverable error states. Add robust input validation, safe filenames for uploads, and graceful handling if an LLM API is unavailable.
Write focused unit tests for the core transcription transformation, speaker renaming logic, and insight extraction, plus one end-to-end happy path test.
Create a detailed README with setup instructions, architecture overview, data location, and backup steps.
Do not add accounts, billing, telemetry, analytics, or a hosted control plane.
Deliberately leave out silent background capture, automatic bot attendance in video meetings, real-time streaming transcription, team workspaces, and enterprise retention controls.
Finish by running the tests and listing the exact commands used to start the application.