Zero-Day Compliance — AI-Powered Policy Enforcement Agent
Link to open source: https://github.com/ManavMalhotra/zero-day-compliance
Link to Live Project: https://8fsphngkeroeamzh3tp8gg.streamlit.app/
Zero-Day Compliance is a multi-agent AI system that automatically reads regulatory policy documents (PDFs), extracts enforceable rules, maps them to any transaction dataset's schema, executes compliance checks, and generates a detailed executive audit report all in under 5 minutes.
The Problem: Financial institutions spend thousands of hours manually reviewing transaction data against evolving AML/KYC policies. Missed violations can result in millions in fines and regulatory action.
The Solution: A 3-agent pipeline powered by Gemini 3 Flash that turns any compliance PDF into actionable SQL/Pandas queries, dynamically maps them to your real data schema (no hardcoding), and produces a Chief Compliance Officer-ready report with risk scores, financial exposure calculations, top offenders, and developer-replicable queries.
Key Features:
- 🔍 Agent 1 — Policy Interpreter: Extracts enforceable rules from PDF policy documents and generates generic SQL + Pandas queries
- 🗺️ Agent 2 — Schema Mapper: Dynamically maps generic queries to any dataset's actual column names and values using AI-powered semantic matching
- 📊 Agent 3 — Compliance Executor: Executes queries, calculates risk scores and financial exposure, and generates a Streamlit-rendered executive report
- ⚡ Sub-5 minutes end-to-end pipeline (PDF upload → full audit report)
- 📥 Export reports as Markdown or PDF
- 🛡️ Zero-config schema mapping — works on any CSV/database without manual column configuration
Tech Stack: Streamlit · Google Gemini 3 Flash Preview · Pandas · Pydantic · PyPDF2
This build was uploaded as a hackathon project
