
DealFlow AI: Autonomous M&A Due Diligence Agent The Problem Mergers & Acquisitions (M&A) are high-stakes operations where speed is paramount. Lawyers spend hundreds of hours manually reviewing thousands of pages of contracts to find "deal breakers" like Change of Control clauses. This manual process is slow, expensive, and prone to human error due to fatigue. The Solution DealFlow AI is an agentic solution acting as an autonomous legal auditor. By leveraging IBM watsonx Orchestrate, it allows legal teams to audit complex documents simply by chatting with an agent. Instead of reading page-by-page, the lawyer instructs the agent: "Audit this contract." The agent handles the workflow autonomously, returning a risk score, extracting specific clauses, and visualizing the data on a dedicated dashboard. Technology & IBM Integration Our project utilizes a Hybrid Agentic Architecture centered on IBM technology: The Core (Orchestrator): We used IBM watsonx Orchestrate to build the conversational agent. It manages user intent and triggers specific skills based on the request. The Brain (Generative AI): The agent connects to a custom Python Skill via OpenAPI. This skill sends contract text to IBM watsonx.ai, utilizing the ibm/granite-3-3-8b-instruct model. We chose Granite for its superior performance in enterprise and legal contexts. The Interface (Analytics): Structured data is pushed to a Streamlit Dashboard for deep-dive visualization. Key Features Automated Risk Scoring: Instantly rates a contract's risk from 0 (Safe) to 100 (Critical). Clause Extraction: Identifies and quotes "Change of Control" and "Assignment" clauses with RAG-like precision. Agentic Workflow: Fully autonomous loop from file upload to final report generation. Drafting Assistant: The agent can auto-draft legal consent letters based on identified parties How I built it IBM watsonx Orchestrate/IBM Granite 3.3/Python (Flask)/Ngrok/Streamlit
23 Nov 2025