pulijon/Sttcast

Uses WhisperX with CUDA acceleration and Pyannote for automatic speaker diarization, processing audio into timestamped transcripts with semantic search via RAG (Retrieval-Augmented Generation) powered by OpenAI embeddings and FAISS vectors. The three-tier architecture separates transcription jobs (port 8000), RAG inference (port 5500), and vector/database queries (port 8001), enabling independent scaling and supporting both GPU-accelerated and CPU-only processing pipelines. Integrates PostgreSQL for web interface state, Flask/FastAPI for multiple frontends, and offers CLI, web UI, and semantic search interfaces for podcast collections.