How criscotero/nlp-chatdash works
Project type and purpose
• NLP ChatDash is a full-stack, Python-based RAG (Retrieval-Augmented Generation) search and FAQ system.
• It ingests PDF documents, extracts embeddings and topics, builds a vector index, and exposes three user interfaces:
• A REST API (FastAPI)
• A Streamlit dashboard
• A Gradio chat UI
File structure
• Top-level:
• app/ – core application code
• data/ – document inputs and generated artifacts
• Makefile, main.py – orchestration entry points
• requirements.txt, pyproject.toml – dependencies
• app/ subdivides into:
• api/ – FastAPI service
• dashboard/ – Streamlit app
• ui/ – Gradio app
• cli.py – Typer CLI commands (build-index, ingest, topics, eval, etc.)
• features/ – code for embeddings, clustering, topic modeling
• pipeline/ – ingestion and indexing workflows, registry, logging
• rag/ – RAG training, inference, evaluation, synthesis
• utils/ & config.py – shared utilities and configuration
README for additional insights
• Workflows orchestrated via Make targets: setup, train, init-corpus, api, dash, gradio, eval, topics, pipeline
• Data flow:
• PDFs in data/docs/
• make init-corpus → ingest PDFs → deduplicate → generate topics (c-TF-IDF) → compute embeddings → build vector index
• make train → build or retrain FAQ index for RAG
• Interfaces load models/indices at runtime
Elements
• Clear labels (component names and file or module examples)
• Yellow for model training/inference
• Orange for UI layers
Detail advice
• Makefile/CLI kicking off each pipeline step.
• That all three frontends share the same config and artifacts directory via app/config.py.