---
title: Financial Intelligence RAG Pipeline
emoji: 📊
colorFrom: blue
colorTo: green
sdk: streamlit
sdk_version: 1.39.0
app_file: app.py
pinned: false
---

# Financial Intelligence RAG Pipeline

A production-grade Retrieval-Augmented Generation (RAG) system over Apple SEC filings and Morningstar research reports.

## What It Does

Ask natural language questions about Apple's financials and get answers grounded in source documents with full citations.

**Example questions:**
- What was Apple's total net sales in FY2024?
- What are Apple's main risk factors from the 2024 10-K?
- How did the Services segment perform compared to Products?
- What is Apple's gross margin trend over the last 3 years?

## Architecture

```
SEC EDGAR (10-K, 10-Q, 8-K) + Morningstar PDFs
          |
    Docling Processing (HTML/PDF parser)
          |
    HybridChunker (tokenizer-aware segmentation)
          |
    all-MiniLM-L6-v2 Embeddings (384-dim)
          |
    Qdrant Cloud (1,234 vectors)
          |
    Two-Stage Retrieval:
      Dense ANN (50 candidates) → ms-marco Cross-Encoder Reranking (top 8)
          |
    Google Gemini 1.5 Flash (streaming generation)
          |
    4-Layer Guardrails (input / retrieval / output / compliance)
          |
    Streamlit Chat UI
```

## Tech Stack

| Component | Technology |
|-----------|------------|
| Document Processing | Docling |
| Embeddings | sentence-transformers/all-MiniLM-L6-v2 |
| Reranker | cross-encoder/ms-marco-MiniLM-L-6-v2 |
| NLI Verifier | cross-encoder/nli-deberta-v3-small |
| Vector Store | Qdrant Cloud |
| LLM | Google Gemini 2.5 Flash |
| RAG Chain | LangChain LCEL |
| UI | Streamlit |
| CI/CD | GitHub Actions → Hugging Face Spaces |

## Evaluation Results

| Metric | Score |
|--------|-------|
| Retrieval Hit Rate | 100% |
| Context Recall | 93.3% |
| Answer Relevancy | 75.5% |
| Faithfulness | 52.7% |
| Aggregate | 55.7% |

## Local Setup

```bash
# 1. Clone and install
git clone <repo-url>
pip install -r requirements.txt

# 2. Configure environment
cp .env.example .env
# Fill in GOOGLE_API_KEY, QDRANT_URL, QDRANT_API_KEY

# 3. Run the app
streamlit run app.py
```

## Environment Variables

| Variable | Required | Description |
|----------|----------|-------------|
| `GOOGLE_API_KEY` | Yes (for deployment) | Google AI Studio API key |
| `QDRANT_URL` | Yes (for deployment) | Qdrant Cloud cluster URL |
| `QDRANT_API_KEY` | Yes (for deployment) | Qdrant Cloud API key |

Without these set, the app falls back to local Ollama + ChromaDB for development.

## Disclaimer

This application is for informational and educational purposes only. It does not constitute investment advice.