AI Audit Trail Generator

A compliance-ready Streamlit app that creates auditable JSON records for every AI interaction.

Features

• Every interaction logged with full model metadata
• Grounded explanations based solely on observable inputs/outputs
• Governance flags for risk domains (financial, medical, HR, legal) and PII
• Append-only JSONL audit log for compliance review
• Download logs as CSV directly from the sidebar
• Runs entirely with free, local, open-source tools (Ollama)

Quick Start

1. Install dependencies

pip install -r requirements.txt

2. Pull the model via Ollama

ollama pull llama3.1:8b

3. Run the app

streamlit run app.py

4. Run tests

pytest tests/ -v

Project Structure

ai-audit-trail/
├── app.py                    # Streamlit UI
├── src/
│   ├── schemas.py            # Pydantic data models
│   ├── llm_client.py         # Ollama wrapper
│   ├── explanation.py        # Observable-evidence rationale engine
│   ├── governance.py         # Rule-based PII & risk detection
│   └── audit_logger.py       # Append-only JSONL logger
├── tests/
│   └── test_audit_system.py  # Unit tests
├── logs/                     # Created automatically at runtime
│   └── audit_log.jsonl
└── artifacts/explanations/   # AIX360-style JSON artifacts

Audit Event Schema

Each line in logs/audit_log.jsonl is a JSON object with these top-level keys:

Field	Description
audit_id	UUID4, unique per interaction
timestamp_utc	ISO 8601 naive UTC
session_id	8-char hex, resets on "New Session"
model	Provider, model name, temperature, max_tokens
request	System prompt, user prompt, context documents
response	Response text, latency in ms
explanation	Rationale summary, evidence used, artifact path
governance	Risk flags, PII detected, policy status