Dev builds real-time dictation pipeline: AssemblyAI Sync API plus LLM Gateway rewrite
tech__unicorn · x · 2026-10-13
A developer shares the architecture of Flow Dictation, a self-built voice-to-text tool that made dictation their primary input method, claiming a 10x productivity boost. The pipeline: AssemblyAI's Sync API (Universal-3.5, 134ms latency, streaming 16kHz mono PCM over WebSocket) produces a raw transcript, then AssemblyAI's LLM Gateway (hosted Qwen 3.5) strips filler words, fixes grammar and keeps the speaker's actual intent, with data staying on-platform. Clean text is pasted directly into the active window — VS Code, JetBrains, Notion — with no context switching. Key insight: a transcript only gives you words; a useful tool gives you the next thing you actually need.
More from coding & agent
- OBVPM timeline workflow v0.1.4-007 fixes joins, audio sync and multi-shot upscaling for MiniMax H3 — obvpm · 2026-10-13
- What Does a Production-Ready Enterprise AI Agent Stack Actually Look Like? Reddit Seeks Answers — done_action · 2026-10-13
- BGPT MCP Server Lets AI Assistants Search Daily-Updated Scientific Data — modelcontextprotocol · 2026-10-13
- Lowvyn MCP Connector Brings Canadian Retail Price Comparison to AI Assistants — modelcontextprotocol · 2026-10-13
- PDF TO MP3: MCP server turns PDFs into audiobooks on your existing Claude or Codex subscription — petrusenko_max · 2026-10-13
- PilotDeck 2.0: open-source agent workbench that ships full data-analysis dashboards — techNmak · 2026-10-13