Risk Hacktory: Reconstructing an Intelligent Analysis Pipeline for Enterprise Risk Management Using LLM

Risk Hacktory is a risk heuristic analysis pipeline that combines Python data analysis and large language model (LLM) feedback. It enables intelligent assessment of risk data and output of an enriched risk register via Jupyter Notebook. Developed by the Projecting Success Solutions Portal team, this project aims to address the pain points of traditional risk management and provide a technical blueprint and reference case for AI-enhanced risk management.

Traditional risk management faces core challenges: vague risk descriptions, lack of systematic mitigation strategies, low efficiency of manual assessment, and difficulty in extracting actionable insights from historical data. As project scale expands, these pain points amplify exponentially, leading to delayed or even ineffective risk responses.

Risk Hacktory is an entry for Hackathon 26, adopting a dual-engine architecture:

1. Python Data Analysis Layer: Using Jupyter Notebook as the environment, it processes structured risk data via Pandas and NumPy, performing statistical analysis, pattern recognition, and data cleaning;
2. LLM Heuristic Evaluation Layer: Applying domain expert rules, the LLM captures semantic nuances (such as risk chain effects and ambiguous descriptions) and outputs structured feedback. The core innovation is an intelligent analysis system rather than a simple recording tool.

Three-stage workflow:

1. Data Ingestion and Preprocessing: Load Excel/CSV data into DataFrame, complete field mapping, missing value handling, and text cleaning;
2. Heuristic Analysis and LLM Enhancement: Python extracts risk features, while the LLM evaluates risk quality (e.g., whether descriptions are specific, whether mitigation measures address root causes, etc.);
3. Output Generation: Annotated spreadsheet (adapted to existing processes) + summary dataset (high-level overview).

Application scenarios include:

• Large-scale engineering projects: Handling hundreds of risk items to reduce manual omissions;
• Enterprise compliance management: Screening risk descriptions in policy documents to ensure compliance;
• Project audits: Standardizing evaluation frameworks to reduce subjective bias;
• Risk management training: Demonstrating methods to integrate AI into traditional processes.

Technical Insights: LLM evolves into an intelligent component of the data pipeline, and human-machine collaboration (Python handles structured data/computations, LLM handles semantics) achieves synergistic effects; Limitations and Outlook: Need to address data privacy (local deployment/desensitization), domain adaptation (configurable rules), and interpretability (reasoning traceability). This project provides a blueprint for AI-enhanced risk management.

Risk Hacktory represents a new paradigm: AI amplifies the judgment of human experts, human-machine collaboration is a feasible path to intelligent risk management, and it serves as a reference case for enterprises exploring AI applications.