How mujtaba-a-khan/call-analytics-system works
Project type and purpose
• This is a full-stack, locally hosted call analytics system.
• Purpose: ingest call audio and CSV transcripts, transcribe speech, label calls, index and semantically search them, and present dashboards and Q&A via a Streamlit UI.
File structure
• Top-level directories:
• config/ (TOML config for audio, models, rules, vectorstore)
• src/
– core/ (audio & CSV processing, labeling, storage)
– ml/ (Whisper STT, embeddings, LLM interface)
– vectordb/ (ChromaDB client, indexer, retriever)
– analysis/ (filters, aggregations, semantic search, query interpreter)
– ui/ (Streamlit app, pages: upload, dashboard, analysis, Q&A; UI components)
– utils/ (logging, file I/O, text utils, validators)
– cli.py (command-line entry points)
• data/ (uploads, processed, cache, vectorstore storage)
• models/ (downloaded Whisper models)
• scripts/ (environment setup, model download, index rebuild)
• tests/
• Configuration: pyproject.toml, config/*.toml
• Deployment/CI: Jenkinsfile, GitHub workflows (implied), Docker not present but can be added.
README for insights
• Tech stack: Python 3.13, Streamlit, ChromaDB, OpenAI Whisper, sentence-transformers, FFmpeg
• Features: audio transcription, CSV ingest, intelligent labeling, semantic search, NL Q&A, dashboards.
• Data flow: upload → process (Whisper STT / CSV) → label → embed → store in ChromaDB → query & retrieve → render in UI.