Technology & Deployment
Open-Source Foundation
- Transcription: Faster-Whisper or OpenAI Whisper (quantized for speed).
- Analysis: Quantized open-source LLMs for entity extraction, inconsistency detection, and summarization.
- Frontend: Streamlit-based local dashboard for case management, audio upload, and chat interface.
- RAG Pipeline: Voice → Text → Embeddings → Semantic Search → LLM Generation.
Recommended Hardware (Pilot)
- Apple MacBook Pro M4 Max (128 GB unified memory, 40-core GPU)
- Alternative: Dedicated air-gapped mini-PC with NVIDIA A6000/A100-class GPU for multi-station use.
- Performance: Real-time transcription, full analysis in <10 minutes per interview.
- No internet required after initial setup.
Deployment Model
- One-time hardware + software bundle.
- Optional annual support/maintenance.
- Pilot available (5 cases, logged time savings).