EchoVault

A blazingly fast, completely private, and highly interactive real-time transcription service powered by Local AI and Next.js.

Modern Web Control Dashboard

EchoVault Web Dashboard

System Architecture

Built entirely on cutting-edge stacks to isolate the insanely heavy, GPU-accelerated Local AI pipeline from the lightweight, responsive React Web Dashboard interface.

The Technology Stack

EchoVault bridges state-of-the-art Python Data Science with Modern Web App Frameworks.

AI Audio Pipeline (Python)

  • Faster-Whisper: GPU-accelerated CTranslate2 inference for real-time speech-to-text.
  • SpeechBrain ECAPA-TDNN: Zero-shot biometric speaker diarization and voice enrollment indexing.
  • Sounddevice & Librosa: Captures native raw hardware audio and resamples on-the-fly dynamically.

Backend Server (FastAPI)

  • FastAPI & Uvicorn: High-performance async REST APIs wrapper around the heavily multithreaded machine learning pipeline.
  • WebSockets: Streams sub-millisecond raw volume spikes, system hardware statistics, and live UI loading states to the browser.
  • Telemetry: Exposes PSUtil and NVML directly mapped to system watermarks in the React Web dashboard.

Frontend Web (Next.js 15)

  • React 19 & Tailwind CSS: Radix-UI/Shadcn polished dashboards featuring micro-interactions and dark modes.
  • Real-Time Connectivity: Dual-channel communication bridging SSE (Server-Sent Events) from MongoDB and pure WebSockets from FastAPI.
  • Stateless LLM Integration: Directly streams the backend's Qwen LLM output via HTTP POST generation calls.

AI Meeting Assistant & Summarizations

Say goodbye to lost focus! The moment you hit "Stop Transcription," EchoVault seamlessly purges your VRAM, wipes the Whisper core, and natively invokes Qwen2.5-0.5B-Instruct via HuggingFace Transformers.

  • Stateless Memory pulling historic logs straight out of Next.js MongoDB
  • Offline & Private—no user recording data is passed to external vendor APIs
  • Interactive Chat capabilities connected to your specific conversational Contexts