🤖 FinRobot: Integrated Financial Analyst

FinRobot is a production-grade, agentic AI system that combines advanced reasoning with autonomous financial data retrieval.

It integrates two powerful systems:

1. Agentic Reasoning Core: A structured cognitive pipeline (Planner → Thinker → Verifier) ensuring explainable and hallucinations-free answers.
2. Autonomous RAG Engine: A control plane that automatically fetches, indexes, and maintains fresh financial data (SEC filings, Stock Prices, News) from the web.

---

🚀 Key Features

• Autonomous Data Acquisition: Automatically detects tikers (e.g., "$AAPL") and fetches missing data (10-K, Prices) before answering.
• Verifiable Reasoning: Every answer goes through a self-verification loop to check for factual accuracy and compliance.
• Long-Term Memory: Remembers user preferences, risk tolerance, and past interactions.
• Hybrid Retrieval: Combines structured financial data with unstructured semantic search over filings.

---

🗂️ Project Structure

FinRobot_Integrated/
│
├── chatbot_ui.py              # Main Entry Point (Streamlit UI)
├── requirements.txt           # Integrated Dependencies
├── .env                       # API Keys Configuration
│
├── agent/                     # Agentic Cognitive Layer
│   ├── meta_agent.py          # Orchestrator
│   ├── planner.py             # Task Decomposition
│   ├── thinker.py             # Reasoning & Synthesis
│   └── verifier.py            # Safety & Fact Checking
│
├── rag_engine/                # Data & Retrieval Engine
│   ├── src/
│   │   ├── control_plane/     # Data Lifecycle (Fetch/Index)
│   │   ├── inference_plane/   # Read-Only Retrieval
│   │   └── orchestrate.py     # RAG Entry Point
│
├── retrieval/                 # Integration Layer
│   └── pinecone_client.py     # Adapts RAG Engine for the Agent
│
├── memory/                    # User Context & Long-term Memory
├── prompts/                   # System Prompts
└── config/                    # Settings & Rules

---

🛠️ Installation

1. Clone & Enter Directory:

   cd FinRobot_Integrated

2. Create Environment:

   python -m venv .venv
   source .venv/bin/activate  # Windows: .venv\Scripts\activate

3. Install Dependencies:

   pip install -r requirements.txt

4. Configure Environment: Create a .env file with the following keys:

   # LLM Providers
   OPENAI_API_KEY=sk-...  (or GEMINI_API_KEY)
   
   # Vector Database
   PINECONE_API_KEY=...
   PINECONE_INDEX_NAME=financial-analysis  (Ensure this index exists)
   PINECONE_ENV=us-east-1
   
   # RAG Data Sources
   HUGGING_FACE_API_KEY=hf_...  (For embeddings)
   NEWSAPI_KEY=...              (Optional: For news)

---

▶️ Usage

Start the application:

streamlit run chatbot_ui.py

Example Workflow

1. Login: Enter a User ID (e.g., "analyst_01").
2. Profile: Set your risk tolerance and explanation depth.
3. Ask: "Analyze the risk factors for Apple based on their latest 10-K."
4. Agent Action:
   • Identifies "AAPL".
   • RAG Engine: Checks if 2025 10-K is indexed. If not, downloads from SEC EDGAR and segments it.
   • Retrieval: Finds relevant risk sections.
   • Thinker: Synthesizes an answer citing specific sections.
   • Verifier: Double-checks the claims against the retrieved text.
   • Response: Delivers the final answer to you.

---

🧠 Architecture Details

The Agent (Reasoning)

The agent avoids "black box" generation by splitting the process:

• Planner: "I need to find Apple's 10-K and look for 'Risk Factors'."
• Thinker: Executes the retrieval and drafts a response.
• Verifier: "The draft claims revenue grew 5%, but the text says 3%. Correction needed."

The RAG Engine (Data)

A "Two-Plane" architecture ensures data integrity:

• Control Plane: Write-Only. Ensures data is fresh. Mirrors local disk to Pinecone.
• Inference Plane: Read-Only. Fast semantic search for the Agent.

---

📄 License

MIT License