SolScribe
GPU-accelerated audio transcription with speaker diarization, AI-powered meeting briefs, and privacy-first processing. Built from the ground up in Go and React.
See It in Action
Dashboard with live transcription progress
Auto-recording with waveform visualization
AI-powered instant meeting briefs with summaries and action items
From Side Project to Full Platform
The Origin
SolScribe started as a personal tool for transcribing meeting recordings on my homelab. I was running WhisperX in Docker and piping results through n8n workflows. The manual pipeline worked but was fragile, so I built a proper application around it.
The Architecture
Go backend with GORM on Neon Postgres. React frontend with real-time progress via WebSocket. WhisperX handles transcription with CUDA acceleration, and PyAnnote provides speaker diarization. The entire stack runs in Docker containers.
Key Engineering Decisions
Chose Go over Node.js for the backend because of its native concurrency model, which is critical when managing GPU-bound transcription jobs. Storage started on SQLite and migrated to Neon Postgres as the platform outgrew single-machine deployments. Browser-based recording with MediaRecorder API for zero-install capture.
AI Integration
LLM-powered features include automatic meeting summaries, key point extraction, action item parsing, and conversational chat over transcripts. The AI layer connects to any OpenAI-compatible endpoint, supporting both cloud and local LLMs.
Built with
See the Full Product
Visit the marketing site for features, pricing, and documentation.