Architecture

Built for speed.
Designed for thought.

Audiomap chains Whisper, LLaMA 3.3, and Groq's inference engine to produce mind maps that feel instantaneous — typically under one second end-to-end.

Request Pipeline

Voice / Text
User input
Whisper v3
Transcription
LLaMA 3.3
Structuring
800+ tok/s
Mermaid Parser
Conversion
React Flow
Rendering

Technology Stack

Groq LPU Inference800+ tok/s

Groq's Language Processing Unit delivers sub-second LLM inference — making Audiomap feel instant rather than waiting on a remote API.

OpenAI Whisper v3Multilingual

Best-in-class automatic speech recognition. Handles Arabic, English, and mixed input with near-human accuracy on Groq's infrastructure.

LLaMA 3.3 70BMeta AI

70-billion parameter open-weights model. Understands complex conceptual relationships and produces structured Mermaid.js mindmap syntax reliably.

Next.js 16 App RouterServerless

API routes keep sensitive keys off the client. Edge-ready for global deployment. Turbopack for fast local development.

React FlowCanvas

Interactive node-based canvas. Supports drag, zoom, pan, custom node types, and minimap navigation out of the box.

IndexedDB via idbPersistence

Maps are saved locally using IndexedDB — no backend, no account required. Instant save, instant retrieval, fully offline capable.

Open source on GitHub

100% open source. Read the code, contribute, or fork it.

mohasbks / Audiomap