Risk Detector

Minimal skeleton of dynamic risk classification chat assistant.

Installation

pip install -r requirements.txt
# or install the package
pip install .

CLI usage

The project exposes a command line interface providing helpers to manage rules and run the interactive dialog.

annex4risk-detector --help

Typical workflow:

# create database tables
annex4risk-detector init_db --db-url sqlite:///risk.db

# load YAML rules into the database
annex4risk-detector sync_rules --db-url sqlite:///risk.db --dir path/to/rules

# start an interactive classification chat
annex4risk-detector chat --db-url sqlite:///risk.db

See risk-detector --help for all available commands.

Demo script

For a quick demonstration with a temporary SQLite database and pre-filled answers that exercise all risk levels, run:

python demo_script.py

The script loads sample rules from test_rules/risk and shows how the dialog flows for scenarios ending in the prohibited, high_risk, not_high_risk (including Art. 6(3) derogation), and out_of_scope categories.

Serverless deployment

When deploying the Lambda handler, use a database reachable from the serverless environment (for example, an Amazon RDS PostgreSQL instance). Expose the connection string via the DB_URL environment variable – handler.py reads this variable and passes it to DialogOrchestrator.

Before deployment initialise the database and load the rules:

annex4risk-detector init_db --db-url "$DB_URL"
annex4risk-detector sync_rules --db-url "$DB_URL" --dir path/to/rules

These commands create tables including chat_answers and risk_outcomes which persist every answer and classification result.