PolicyGuard

AI-powered compliance agent for APAC fintech teams — built for Gen AI Academy APAC · Track 1 · Banking & Financial Services

PolicyGuard monitors regulatory feeds from MAS, HKMA, and ACRA, scores compliance obligations, generates structured action plans, and surfaces alerts — all in a single deployable service with an embedded dashboard.

---

The Problem

Fintech teams in APAC face three compounding challenges:

Challenge	Impact
Frequent regulatory updates from MAS/HKMA/ACRA	Teams must monitor multiple regulators simultaneously
Manual review loops via spreadsheets and email	Weak traceability and inconsistent interpretation
GRC tools are expensive	Smaller fintechs lack affordable structured compliance workflows

---

What PolicyGuard Does

Regulatory Update → Ingest → Assess → Summarize → Plan → Track → Dashboard Alert

1. Ingest — Polls regulatory feeds and captures raw policy inputs
2. Assess — Scores applicability, severity, and enforceability (0–100)
3. Summarize — Converts long-form policy text into concise, actionable guidance
4. Plan — Generates deadline-aware action items and remediation payloads
5. Track — Retains versioned policy history and surfaces alerts in the dashboard

---

Quick Start

1. Clone and set up

git clone https://github.com/om-pramod/PolicyGuard.git
cd PolicyGuard
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

2. Run locally

uvicorn app.main:app --reload

Open http://localhost:8000/ to view the Regulatory Pulse dashboard.

3. Try the APIs

# Run a compliance check
curl -s http://localhost:8000/compliance-check \
  -H 'content-type: application/json' \
  -d @samples/compliance_check_sample.json

# Generate an action plan
curl -s http://localhost:8000/actions/plan \
  -H 'content-type: application/json' \
  -d @samples/action_plan_sample.json

4. Run tests

pytest -q

---

Deploy to Cloud Run

PROJECT_ID=<your-gcp-project> \
REGION=asia-southeast1 \
SERVICE_NAME=policyguard \
bash scripts/deploy_cloud_run.sh

Or use the provided Dockerfile, cloudbuild.yaml, and cloud-run.yaml directly.

---

API Reference

Endpoint	Method	Description
/	GET	Embedded React + Plotly dashboard
/health	GET	Service health check
/compliance-check	POST	Score and assess a regulation
/feed/status	GET	Live MAS/HKMA/ACRA feed status
/alerts/preview	GET	Preview critical alert queue
/alerts/dispatch	POST	Dispatch webhook alert
/actions/plan	POST	Generate prioritized action plan
/policy/diff	POST	Semantic diff between policy versions
/ai/summarize	POST	AI-powered regulation summarizer
/versions	GET	Policy version history

---

AI Summarization Providers

PolicyGuard supports pluggable, free AI providers — no paid API keys required:

Provider	How to use
heuristic	Built-in fallback, always available
ollama	Run local OSS models (Llama, Mistral, etc.)
huggingface	Free-tier HuggingFace Inference API

See docs/free_ai_model_resources.md for setup instructions.

---

Tech Stack

• Backend — FastAPI (Python 3.11), modular routers and services
• Dashboard — Embedded React + Plotly (served from /)
• AI — Pluggable providers: heuristic, Ollama, HuggingFace
• Google OSS — OR-Tools (action prioritization), diff-match-patch (policy diffing)
• Storage — In-memory TTL cache + file-backed JSONL version log (no external DB)
• Security — HTTPS enforcement, per-IP rate limiting, hardened headers, optional Google ID token auth
• Deployment — Docker + Cloud Run (asia-southeast1)

---

Sample Files

File	Purpose
samples/compliance_check_sample.json	Sample request for POST /compliance-check
samples/action_plan_sample.json	Sample request for POST /actions/plan
samples/policy_diff_sample.json	Sample request for POST /policy/diff
samples/ai_summarize_sample.json	Sample request for POST /ai/summarize
samples/grc_action_payload_sample.json	Sample structured GRC remediation output

---

Documentation

Doc	Contents
docs/current_state_guide.md	Plain-English walkthrough — how it's built and how to demo
docs/policyguard_development_plan.md	Full architecture, Cloud Run, timeline, and KPI plan
docs/policyguard_sg_fintech_playbook.md	Singapore fintech-focused implementation details
docs/policyguard_detailed_report.md	Step-by-step execution report for compliance teams
docs/google_open_source_extensions.md	OR-Tools and diff-match-patch integration notes
docs/security_hardening_google_cloud.md	Security middleware and Google Identity setup
docs/next_steps.md	Execution checklist from local to production

---

Project Status

• FastAPI backend with all required endpoints (app/)
• Embedded React + Plotly dashboard (static/index.html)
• Docker + Cloud Run deployment configs
• Tests for core scoring and remediation logic (tests/)
• Sample payloads for all major endpoints (samples/)
• Google OSS extensions (OR-Tools, diff-match-patch)
• Pluggable free AI providers (heuristic, Ollama, HuggingFace)

---

About

Built by Omkar Hankare for the Gen AI Academy APAC Edition hackathon.

• Track 1 — Build and deploy AI agents using Gemini, ADK, and Cloud Run
• Theme — Banking & Financial Services · Positive social change through safer fintech compliance
• Region focus — Singapore, Hong Kong, India (MAS / HKMA / ACRA regulators)

PolicyGuard helps APAC fintech teams respond faster, document better, and reduce preventable compliance risk.