AI Incident Helper - Automatic Extraction and structured reports

Description:
Takes the raw incident descriptions , extracts the key entities (perpetrators, locations , violations ) , generates a structured incident report 
and let the users submit their Feedback
AI powered agents that takes raw incident descriptions and generate a strcutured incident reports. It saves out the time in manually filling up the data 


Features:

1. Text cleaning
2. LLM driven entity Extraction
3. Full incident report generation
4. Feedback logging
5. Console UI for interactive use

Concepts Applied
1. Multi-Agent System (Sequential)
  - A sequential multi-agent pattern has been used
    - Cleaning Agent : Normalize and tokenize text
    - Extraction Agent : LLM-based entity extraction (perpetrators, locations , violations)
    - Feedback logger

  WorkFlow: Cleaning -> Extraction -> Feedback
2. Sessions and Memory (State + Memory Bank)
  - Session : 
  - Memory :
3. Observability and Agent Evaluation



report_structuring_agent/
│
├── cleaning_agent.py        # Normalization & token cleaning
├── extraction_agent.py      # LLM-based entity extraction
├── indexing_agent.py        # Searchable storage
├── console_app.py           # Interactive CLI app
├── evaluation.py            # Gold test evaluation
├── generated_reports/       # Output incident reports
└── customer_feedback.txt    # Feedback logs


Setup Instructions
brew install ollama
ollama pull llama3.2


Usage Instructions
1. Run the following command from the source code location folder 
python3 console_app.py

2. Provide the incident description
Enter incident description: John shouting at desk
...
Entities:
  Perpetrators : John
  Locations    : desk
  Violations   : shouting


Feedback Handling
After each report the user can mark wrong extraction fields.
Feedback is apended to customer_feedback.txt for auditing, model improvement or future fine tuning

Limitations
1. Extractor accuracy depends on LLM quality
2. No UI yet (console only)