Ramble
A voice-to-structured-card MVP scaffold that records or uploads audio, creates saved cards, supports search/edit/copy/delete, and keeps provider boundaries clear.
What it is
Ramble focuses on the moment after someone talks into their phone and wants a useful artifact, not a raw transcript.
The v1 promise is deliberately narrow: turn a short spoken mess into a note, task, or message card that can be saved, copied, edited, searched, or deleted.
The loop it has to make believable
- 01
Locked v1 to three modes so the product does not sprawl into a broad assistant.
- 02
Built the backend in Go around idempotent audio uploads, recordings, status polling, cards, transforms, search, edit, copy, and delete.
How it’s built
The local implementation proves the architecture before paid providers are introduced. The Go API owns recordings, cards, status, search, and provider orchestration.
Mock providers make the loop runnable, while the Sarvam adapter shows how real speech transcription slots in. The unfinished parts are also explicit: no production auth, no background worker, no semantic search, no production LLM structuring, and no durable user-audio storage.
Architecture
- Capture
- Record in browser or upload audio, then create a recording through multipart POST
- Backend
- Go net/http API for recordings, status polling, cards, transforms, search, edit, and delete
- Processing
- Mock transcription/structuring by default; optional Sarvam Saaras v3 STT through provider config
- Boundary
- Per-browser demo user ids, inline processing, local/ephemeral audio storage, and SQL text search are not beta-ready auth/queue/search
What has to hold up
- Kept transcription and structuring swappable through provider interfaces, with Sarvam STT available behind config.
- Documented the follow-up path for production auth, queues/retries, cost controls, privacy/retention controls, durable object storage, and better structuring providers.
How I would present it
I would be explicit that this is a scaffold, not a launched SaaS. The value is the product boundary, Go/Postgres model, idempotent recording creation, provider seam, and full local voice-to-card loop.
The next serious step is quality evaluation for real transcription and structured-output providers.
Proof
VerifyRepository, checks, stack, status
- 01
Start with the project status and available evidence.
- 02
Read the architecture and edge-case sections.
- 03
Follow the documented proof path: README, product docs, Go build/test scripts, provider seam.
- Role
- Product scope, Go API, data model, provider seams, frontend loop
- Scope
- Voice note to structured card scaffold
- Proof
- README, product docs, Go build/test scripts, provider seam
- Stack
- Go, Next.js, React, TypeScript, PostgreSQL, pgx, Tailwind CSS
- Modes
- Note/Task/Message
- Backend
- Go API
- STT
- Mock + Sarvam
- Storage
- Postgres + disk