📊 Financial RAG Analyst: SEC 10-K Insight Engine

A production-ready Retrieval-Augmented Generation (RAG) application that analyzes complex financial documents (SEC 10-K filings) with high precision. Built with LlamaIndex, Qdrant, and Groq (Llama 3), deployed on Streamlit Cloud.

Now features Enterprise Observability: Includes a secure Admin Dashboard to track hallucinations, monitor latency, and gather user feedback (RLHF).

🚀 Live App

Click here to try the Live App

📸 Interface Preview

1. The Analyst (Chat Interface)

       

2. The Observer (Admin Dashboard)

Monitor system performance and user feedback in real-time.

3.Mind Map

Overall structure

💡 What It Does

Financial reports like SEC 10-K filings are hundreds of pages long, filled with complex tables and dense legal text. Standard LLMs often hallucinate specific numbers when asked to compare data across these documents.

Financial RAG Analyst solves this by:

1. Ingesting raw PDFs using LlamaParse to accurately extract table structures.
2. Indexing data into Qdrant Cloud using semantic vector embeddings.
3. Retrieving the exact context (e.g., page 52, Table 3) before answering.
4. Synthesizing a hallucination-free answer using Llama 3 via Groq.
5. Monitoring quality via a feedback loop (Thumbs Up/Down) stored in a secure log.

🏗️ System Architecture

graph LR
    A[PDF 10-K Filings] -->|LlamaParse| B(Structured Markdown)
    B -->|Chunking| C[Text Chunks]
    C -->|BGE Embeddings| D[(Qdrant Cloud DB)]
    E[User Question] -->|Semantic Search| D
    D -->|Retrieve Top-5 Contexts| F[Prompt Construction]
    F -->|Groq Llama 3| G[Final Answer]
    G -->|Streamlit UI| H[User Screen]
    H -->|Thumbs Up/Down| I[Logger System]
    I -->|Analytics| J[Admin Dashboard]

Loading

⚡ Key Features

Multi-Document Analysis: Compare metrics across Apple, Microsoft, and Google simultaneously.

Table-Aware Parsing: accurately reads financial tables that confuse standard PDF parsers.

Source-Grounded Truth: Answers are strictly based on the provided filings, eliminating "model bias."

Cloud-Native: Vector data persists in Qdrant Cloud; App runs on Streamlit Community Cloud.

AI Observability: Built-in logging system to track user queries and flag hallucinations.

Zero-Cost Inference: Powered by Groq's free tier, making the app blazing fast and free to run.

🛠️ Tech Stack

Framework: LlamaIndex

Frontend: Streamlit

Vector Database: Qdrant

Parsing: LlamaParse

DevOps: Docker, GitHub Actions (CI/CD)

LLM: Meta Llama 3 (via Groq API)

🏎️ Quick Start (Run Locally)

1. Clone the Repo

Bash

git clone [https://github.com/SIYAM1809/Financial-RAG-Analyst.git](https://github.com/SIYAM1809/Financial-RAG-Analyst.git)
cd Financial-RAG-Analyst

2. Install Dependencies

Bash

pip install -r requirements.txt

3. Set Up Secrets Create a .streamlit/secrets.toml file (or set environment variables) with your keys:

Ini, TOML

# .streamlit/secrets.toml
GROQ_API_KEY = "gsk_..."
QDRANT_URL = "[https://your-cluster-url.qdrant.io](https://your-cluster-url.qdrant.io)"
QDRANT_API_KEY = "your-qdrant-key"
ADMIN_PASSWORD = "admin123"

4. Run the App

Bash

streamlit run app.py

🧪 Demo Questions to Try

"Compare the cloud revenue growth rate of Microsoft Azure vs. Google Cloud for 2024."

"What are the top risk factors listed by Apple regarding AI regulation?"

"How much capital did Microsoft return to shareholders via buybacks in 2024?"

📂 Repository Structure

├── app.py                 # Main Streamlit application
├── logger.py              # Logging system for observability
├── pages/
│   └── admin_dashboard.py # Secure analytics dashboard
├── requirements.txt       # Python dependencies
├── Dockerfile             # Container configuration
├── .github/workflows/     # CI/CD Automated Testing
├── ingest_data.py         # Script to parse PDFs & upload to Qdrant
├── sec-edgar-filings/     # Raw PDF documents (Not included in repo)
└── README.md              # Documentation

🤝 Contributing

Pull requests are welcome! For major changes, please open an issue first to discuss what you would like to change.

📄 License

MIT