sunkistCAT/standard_chartered_rag_demo

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

📊 Standard Chartered RAG Demo

Interact with financial reports or news articles using Google Gemini and Retrieval-Augmented Generation (RAG). Upload a document or paste text, then chat with Gemini for intelligent, concise answers.

✨ Features

• —Upload financial documents (.pdf, .txt, .docx)
• —Paste raw financial news or text
• —Ask questions using Gemini via google-genai
• —Automatic summary and character count for uploaded/pasted content
• —Context-aware answers using RAG (PGVector + Gemini)
• —"Clear Chat" resets all fields and context
• —Built with Gradio UI

📂 Folder Structure

standard_chartered_rag_demo/
├── app.py
├── handlers/
│   ├── chat_handler.py
│   ├── file_handler.py
│   ├── text_handler.py
├── utils/
│   ├── context_utils.py
│   ├── db_service.py
│   ├── gemini_service.py
├── .env
├── requirements.txt
├── README.md

🚀 Setup & Run

1. Install dependencies

pip install -r requirements.txt

2. Configure environment

Create a .env file in the root directory with your API keys and database URL:

GEMINI_API_KEY=your-google-api-key
MODEL_3=gemini-2.5-flash-lite-preview-06-17
EMBEDDING_MODEL=gemini-embedding-001
SUPABASE_DB_URL=your-postgres-connection-string

3. Start the app

python app.py

Open the link provided in the terminal.

🧠 Example Questions

After uploading or pasting content, try:

• —What are the main risks mentioned?
• —Summarize the Q1 financial highlights.
• —What are the revenue growth drivers?
• —Is there any forecast for the next quarter?
• —What are the top risks mentioned in this report?
• —Summarize the company’s performance this quarter.
• —How did revenue change from the previous period?
• —What risks are mentioned in this document?
• —Summarize the financial highlights.
• —What is the outlook for the next quarter?

🧪 Notes

• —Context is truncated to 30,000 characters to avoid overloading Gemini.
• —Both file and text inputs have their own summaries and character counters.
• —Uses PGVector for vector similarity search.
• —Gradio UI for easy interaction.

📌 Tech Stack

• —Google Gemini (google-genai)
• —Gradio
• —Python (3.11+)
• —PyMuPDF (fitz) for PDF parsing
• —python-docx for Word files
• —SQLAlchemy + PGVector for vector database

💡 Future Improvements

• —Authentication, permission handling (use case etc.)
• —RAG optimization
• —Late chunking
• —Query transformation
• —Use local model for compliance, and save cost
• —Check document classification:
• —Allow PUBLIC, INTERNAL
• —Block RESTRICTED, HIGHLY RESTRICTED, or unclassified
• —Model evaluation. # TODO: reference Warden
• —Groundness
• —Context relevance (when using RAG)
• —Answer relevance
• —Model moderation
• —Fairness and bad actor
• —prompt jailbreaking
• —harmful content classifier
• —Feedback service
• —Allow user to submit feedback
• —If a user copied a LLM response
• —monolithic -> microservices
• —Now: UI, business logic, data are all in one place