Automated Forensic Accounting & Management Integrity Analysis System

ForensicValue AI is a multi-agent system designed to detect accounting irregularities, governance risks, and "value traps" in Indian listed companies (NSE/BSE).

• Multi-Agent Architecture:
  • Forensic Agent: Analyzes 10-year financials & annual report text for anomalies.
  • Management Agent: Investigates promoter pledging, board independence, and related parties.
  • RPT Agent: Scrutinizes Related Party Transactions for fund siphoning.
  • Market Intelligence Agent: Searches web/social media for fraud allegations & sentiment.
  • Narrative Agent: Synthesizes all findings into a "Forensic Detective Story".
• Deep Document Analysis: Downloads and parses Annual Reports & Concall Transcripts for rare insights.
• Self-Healing Pipeline: Robust error handling with json_repair and auto-retries.
• Interactive Dashboard: Streamlit-based UI for reports and batch processing.

1. Clone the repository:

   git clone https://github.com/sanjibani/forensic-value-ai.git
   cd forensic-value-ai

2. Install dependencies:

   pip install -r requirements.txt

3. Set up environment: Create a .env file (see .env.example):

   # Required
   GOOGLE_API_KEY=your_gemini_key
   # Optional
   ANTHROPIC_API_KEY=your_claude_key
   OPENAI_API_KEY=your_openai_key

4. Run the Dashboard:

   streamlit run dashboard.py

1. Push this code to your GitHub repository.
2. Go to share.streamlit.io.
3. Connect your GitHub account.
4. Select the forensic-value-ai repo.
5. Set "Main file path" to dashboard.py.
6. Add your API keys (GOOGLE_API_KEY, etc.) in the Advanced Settings -> Secrets section.
7. Click Deploy.

1. Build & Run:

   docker-compose up --build

2. Access at http://localhost:8501.

The system uses LangGraph to orchestrate agents: Fetcher -> Forensic/Mgmt/RPT/Market (Parallel) -> Aggregator -> Critic -> Narrative -> Report.

MIT License.