How kvj-harsha/rag-qna-chatbot works
Project type and purpose
• This is a single-repo, Streamlit-based “retrieval-augmented” Q&A chatbot.
• It provides two UI pages—one for ingesting documents into a Chroma vector store, and one for conversational Q&A over that store using OpenAI’s models.
File structure
• Top level:
• main.py – entry point registering Streamlit pages
• chroma_services.py – vector-store ingestion & query logic
• genai_services.py – OpenAI embedding & chat-completion calls
• data.txt – sample data for ingestion
• template.env – environment variable placeholders (API keys, etc.)
• requirements.txt – Python dependencies
• .gitignore, batch scripts – dev setup files
• pages/
• ingest_page.py – UI for uploading/ingesting documents
• chatbot_page.py– UI for interactive Q&A
• No separate backend server: UI and business logic live in the same Streamlit process.
Insights from README
• Technologies in use:
• Streamlit – UI framework
• OpenAI API – embeddings & chat completions
• tiktoken – token counting
• ChromaDB – local vector database
• markitdown – Markdown rendering support
a. Main components
• Streamlit frontend
– ingest_page
– chatbot_page
• Service layer
– chroma_services.py (ingestion, semantic search)
– genai_services.py (make embeddings, chat completions)
• Data layer
– ChromaDB local vector store
– source docs (data.txt or uploaded files)
• External service
– OpenAI API
• Configuration
– .env / template.env for API keys
b. Relationships & interactions
1. User interacts with Streamlit UI
2. Ingest page calls chroma_services.ingest(docs) → stores embeddings in ChromaDB
3. Chat page sends user query →
i. chroma_services.query(query) → returns relevant context chunks
ii. genai_services.chat_completion(system_prompt + context + question) → returns answer
4. Response rendered back in UI
c. Architectural patterns
• Monolithic Streamlit app with clear separation of concerns:
• Presentation layer (pages/)
• Service layer (…_services.py)
• Data persistence (ChromaDB)
• Retrieval-augmented generation (RAG) pattern
d. Key technologies
• Streamlit, ChromaDB Python client, OpenAI Python SDK, tiktoken
Final advice
• RAG flow end-to-end, from user input through retrieval and LLM inference back to the UI.
• Avoid unnecessary detail: focus on those modules that appear in the file tree.
• This level of granularity ensures any new engineer or stakeholder sees exactly how UI, services, data, and external APIs interconnect.