🏛️AI Hackathon Accelerators for Government Services

---

📋 Executive Summary

What Are Accelerators?

Accelerators are a Microsoft term for complete, working applications that serve as production-ready starting points for your own solutions. Rather than building from scratch, accelerators let you:

• ✅ Learn faster - See real-world implementations of AI patterns and best practices
• ✅ Build securely - Start with security, compliance, and responsible AI already built in
• ✅ Ship sooner - Extend working code instead of writing boilerplate from zero
• ✅ Reduce risk - Leverage tested, validated architectures designed for government use

Think of accelerators as fully-functional blueprints - they work out of the box, but are designed for you to customize and extend for your specific agency needs.

This Repository

This repository contains 7 AI accelerators designed to transform how NY State government agencies serve constituents. Each accelerator is a complete application built with Microsoft Azure AI services and the Semantic Kernel framework, demonstrating practical AI solutions for:

Challenge	Accelerator Solution
📞 Citizens can't find answers	AI chatbot with citations
📄 Document processing backlogs	Automated OCR & validation
🚨 Emergency coordination gaps	Multi-agent planning system
📋 Policy compliance burden	Automated document review
🔍 Siloed agency knowledge	Cross-agency secure search
🏙️ NYC citizen services (.NET)	RAG-powered .NET chatbot
🤖 NYC citizen services (Python)	RAG-powered Python chatbot

All accelerators comply with NY State's LOADinG Act and RAISE Act requirements for transparent, accountable AI in government.

---

📊 Revision Matrix

Version	Date	Changes	Status
2.2.0	Jan 13, 2026	Added Python Virtual Citizen Assistant, 267 tests	✅ Current
2.1.0	Jan 12, 2026	Added .NET Virtual Citizen Assistant, 265 tests	✅ Stable
2.0.0	Jan 12, 2026	Production release with 5 accelerators	✅ Stable
1.5.0	Jan 10, 2026	Added Inter-Agency Knowledge Hub accelerator	✅ Stable
1.4.0	Jan 9, 2026	Added Policy Compliance Checker accelerator	✅ Stable
1.3.0	Jan 8, 2026	Added Emergency Response Agent accelerator	✅ Stable
1.2.0	Jan 7, 2026	Added Document Eligibility Agent accelerator	✅ Stable
1.1.0	Jan 6, 2026	Added Constituent Services Agent accelerator	✅ Stable
1.0.0	Jan 5, 2026	Initial repository setup with shared infrastructure	✅ Stable

---

🚀 The 7 AI Accelerators

1️⃣ Constituent Services Agent

🎯 Purpose: AI-powered chatbot answering citizen questions about NY State services

✨ Key Features:

• 💬 Natural language Q&A about SNAP benefits, driver's licenses, unemployment, Medicaid
• 📚 Citation-backed responses with source documents
• 📊 Confidence scoring and human escalation when uncertain
• 🌍 Multi-language support (English, Spanish, Chinese, Arabic, Russian, Korean, Haitian Creole, Bengali)
• ♿ WCAG 2.1 AA accessible web interface

🛠️ Tech Stack: Azure AI Foundry + Foundry IQ + Semantic Kernel + Flask

▶️ Demo Command:

cd Constituent-Services-Agent
pip install -r requirements.txt
python demo.py

💡 Sample Queries:

• "How do I apply for SNAP benefits?"
• "How do I renew my driver's license?"
• "Am I eligible for Medicaid?"

---

2️⃣ Document Eligibility Agent

🎯 Purpose: Automated processing of eligibility documents (W-2s, pay stubs, utility bills)

✨ Key Features:

• 📧 Email inbox monitoring for document submissions
• 🔍 OCR and intelligent data extraction using Azure Document Intelligence
• 📊 Confidence scoring for all extracted fields
• 🔒 PII detection and automatic masking
• ✅ Validation rules (document age, completeness)
• 📋 Case routing and workload distribution

🛠️ Tech Stack: Azure Document Intelligence + Microsoft Graph + Semantic Kernel + Flask

▶️ Demo Command:

cd Document-Eligibility-Agent
pip install -r requirements.txt
python demo.py

📄 Supported Document Types:

Document	Fields Extracted
W-2 Forms	Wages, employer, tax year
Pay Stubs	Gross pay, period, date
Utility Bills	Provider, address, date
Bank Statements	Institution, balance
Driver's Licenses	Name, DOB, expiration
Birth Certificates	Name, DOB, parents
Lease Agreements	Landlord, address, rent

---

3️⃣ Emergency Response Agent

🎯 Purpose: Multi-agent system for emergency response planning and coordination

✨ Key Features:

• 🌀 Emergency scenario simulation (hurricane, fire, flood, winter storm, public health, earthquake)
• 🌤️ Real-time weather integration
• 🏛️ Multi-agency resource coordination (FDNY, NYPD, OEM, DOT, MTA)
• 🚗 Evacuation route planning with bottleneck analysis
• 📜 Historical incident analysis for lessons learned
• ⏱️ Response plans with timeline milestones

🛠️ Tech Stack: Semantic Kernel + Azure AI Foundry + Weather APIs + Multi-Agent Orchestration

🚨 Supported Emergency Types:

Type	Lead Agency	Key Resources
🌀 Hurricane	OEM	Evacuation, shelters
🔥 Fire	FDNY	Firefighters, equipment
🌊 Flooding	OEM	Pumps, rescue boats
❄️ Winter Storm	DOT	Plows, salt trucks
🏥 Public Health	DOH	Healthcare workers, vaccines
🏚️ Earthquake	OEM	Search & rescue teams
⚡ Infrastructure	Utilities	Emergency generators

---

4️⃣ Policy Compliance Checker

🎯 Purpose: Automated review of policy documents against compliance rules

✨ Key Features:

• 📄 Document parsing (PDF, DOCX, Markdown)
• 🔍 Rule-based compliance checking with regex patterns
• ⚠️ Severity categorization (Critical, High, Medium, Low)
• 📊 Compliance scoring (0-100)
• 🤖 AI-powered analysis with Azure OpenAI
• 💡 Detailed recommendations for each violation
• 🔄 Version comparison for policy changes

🛠️ Tech Stack: Azure AI Foundry + Semantic Kernel + Document AI

▶️ Demo Command:

cd Policy-Compliance-Checker
pip install -r requirements.txt
python demo.py

📋 Compliance Categories:

Category	Description	Examples
Data Privacy	PII handling rules	Encryption, retention
Accessibility	WCAG compliance	Alt text, contrast
Security	Security standards	Authentication, logging
Documentation	Policy requirements	Version control, approval

---

5️⃣ Inter-Agency Knowledge Hub

🎯 Purpose: Cross-agency document search with permission-aware results

✨ Key Features:

• 🔍 Unified search across 5+ agency knowledge bases (DMV, DOL, OTDA, DOH, OGS)
• 🔐 Entra ID authentication with role-based access
• 🛡️ Permission-aware result filtering
• 📚 Citation tracking for LOADinG Act compliance
• 🔗 Cross-agency policy cross-references
• 👤 Human-in-the-loop for complex queries
• 📋 7-year audit log retention

🛠️ Tech Stack: Microsoft Foundry + Foundry IQ + Azure AI Search + Entra ID

▶️ Demo Command:

cd Inter-Agency-Knowledge-Hub
pip install -r requirements.txt
python demo.py

🏛️ Supported Agencies:

Agency	Domain	Documents
DMV	Transportation	Licensing, registration
DOL	Labor	Employment, wages
OTDA	Social Services	Benefits, assistance
DOH	Health	Public health, regulations
OGS	General Services	Procurement, facilities

---

6️⃣ Virtual Citizen Assistant (.NET)

🎯 Purpose: RAG-powered AI assistant for NYC government services built with .NET

✨ Key Features:

• 💬 AI chat assistant with source citations
• 🔍 Semantic, keyword, and hybrid search modes
• 📂 Category browser with visual grid layout
• 📄 Document details with print and share
• 🛠️ Data upload utility for Azure AI Search
• 🎨 Bootstrap 5.3 responsive UI

🛠️ Tech Stack: .NET 9 + ASP.NET Core MVC + Semantic Kernel 1.65 + Azure AI Search + Azure OpenAI

▶️ Demo Command:

cd DotNet-Virtual-Citizen-Assistant
dotnet restore
dotnet run --project VirtualCitizenAgent

💡 Sample Features:

• Chat with AI about NYC services
• Search documents semantically
• Browse by service category

---

7️⃣ Virtual Citizen Assistant (Python)

🎯 Purpose: RAG-powered AI assistant for city government services built with Python

✨ Key Features:

• 💬 Natural language Q&A about city services (trash pickup, permits, emergency alerts)
• 🔍 Vector search + keyword search + hybrid search modes
• 🔌 Plugin architecture with Semantic Kernel 1.37
• 📅 Appointment scheduling with mock service
• 📚 Citation-backed responses with source documents
• 🧪 Built-in test framework for validation

🛠️ Tech Stack: Semantic Kernel 1.37 + Azure AI Search + Azure OpenAI + Flask

▶️ Demo Command:

cd Virtual-Citizen-Assistant
pip install -r requirements.txt
python test_setup.py      # Validate setup
python test_plugins.py    # Test plugins
python src/main.py        # Run interactive assistant

💡 Sample Queries:

• "When is my next trash pickup?"
• "How do I apply for a business permit?"
• "Are there any current emergency alerts in my area?"

🔌 Available Plugins:

Plugin	Functions	Purpose
DocumentRetrieval	search_city_services, get_service_by_category	Search city service information
Scheduling	check_availability, scheduling_info, list_schedulable_services	Appointment management

---

🏗️ Technical Architecture

┌─────────────────────────────────────────────────────────────────┐
│                     🖥️ Frontend Layer                           │
│   Flask Web UI  │  REST APIs  │  WCAG 2.1 AA Accessible        │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│                  🤖 AI Orchestration Layer                      │
│   Semantic Kernel  │  Foundry IQ  │  Multi-Agent Patterns      │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│                   ☁️ Azure AI Services                          │
│  Azure OpenAI GPT-4o  │  Document Intelligence  │  AI Search   │
│  Microsoft Graph      │  Translator            │  Entra ID     │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│                      💾 Data Layer                              │
│   SQLite/Azure SQL  │  Blob Storage  │  Vector DBs             │
└─────────────────────────────────────────────────────────────────┘

---

🛡️ Compliance & Responsible AI

📜 NY LOADinG Act Compliance

• ✅ All AI decisions are logged with rationale
• ✅ Human-in-the-loop for benefits determinations
• ✅ Transparent citation of data sources
• ✅ Bias testing across demographic groups

📋 NY RAISE Act Requirements

• ✅ AI assistance clearly disclosed to users
• ✅ Accountability measures for automated decisions
• ✅ Regular auditing and evaluation frameworks
• ✅ Azure AI Evaluation integration for red-teaming

🔒 Security & Privacy

• 🏛️ Azure GCC (Government Community Cloud) compatible
• 🔐 PII detection and automatic masking
• 👤 Role-based access control via Entra ID
• 🔒 Encrypted data at rest and in transit
• 🗑️ 30-day conversation data purge policy

---

📁 Project Structure

newyork/
├── 📂 Constituent-Services-Agent/     # 💬 Citizen Q&A chatbot
│   ├── src/
│   │   ├── agent/                     # AI agent components
│   │   ├── api/                       # Flask routes
│   │   ├── models/                    # Data models
│   │   └── services/                  # Knowledge service
│   ├── demo.py                        # Interactive demo
│   └── requirements.txt
│
├── 📂 Document-Eligibility-Agent/     # 📄 Document processing
│   ├── src/
│   │   ├── agent/                     # Processing agents
│   │   ├── api/                       # REST endpoints
│   │   ├── models/                    # Document models
│   │   └── services/                  # OCR, email, storage
│   ├── demo.py
│   └── sample_documents/
│
├── 📂 Emergency-Response-Agent/       # 🚨 Emergency planning
│   ├── src/
│   │   ├── models/                    # Emergency models
│   │   ├── orchestration/             # Multi-agent coordinator
│   │   └── services/                  # Weather, traffic APIs
│   └── requirements.txt
│
├── 📂 Policy-Compliance-Checker/      # 📋 Compliance checking
│   ├── src/
│   │   ├── models/                    # Compliance models
│   │   ├── services/                  # Rule engine, parsing
│   │   └── api/                       # Flask routes
│   └── requirements.txt
│
├── 📂 Inter-Agency-Knowledge-Hub/     # 🔍 Cross-agency search
│   ├── src/
│   │   ├── models/                    # Search models
│   │   ├── services/                  # Search, auth services
│   │   └── api/                       # Flask routes
│   └── requirements.txt
│
├── 📂 DotNet-Virtual-Citizen-Assistant/  # 🏙️ NYC .NET chatbot
│   ├── VirtualCitizenAgent/           # Main web application
│   │   ├── Controllers/               # MVC and API controllers
│   │   ├── Services/                  # Business logic
│   │   ├── Plugins/                   # Semantic Kernel plugins
│   │   └── Views/                     # Razor views
│   ├── VirtualCitizenAgent.Tests/     # xUnit tests
│   └── AzureSearchUploader/           # Data upload utility
│
├── 📂 Virtual-Citizen-Assistant/        # 🤖 NYC Python chatbot
│   ├── src/
│   │   ├── config/                    # Configuration settings
│   │   ├── models/                    # Data models
│   │   ├── plugins/                   # Semantic Kernel plugins
│   │   └── main.py                    # Main application
│   ├── test_setup.py                  # Setup validation
│   ├── test_plugins.py                # Plugin tests
│   └── requirements.txt
│
├── 📂 docs/                             # 📖 Documentation
│   ├── QUICKSTART.md                    # Quick start guide
│   ├── EVAL_GUIDE.md                    # Evaluation guide
│   └── SPEC_TEMPLATE.md                 # Specification template
│
├── 📂 evaluation/                       # 🧪 AI evaluation framework
│   ├── eval_config.py                   # Evaluation configuration
│   ├── run_evals.py                     # Run evaluations
│   ├── red_team.yaml                    # Red team test config
│   └── test_cases.jsonl                 # Test cases
│
└── 📂 specs/                            # 📋 Feature specifications
    ├── 001-constituent-services-agent/
    ├── 002-document-eligibility-agent/
    ├── 003-emergency-response-agent/
    ├── 004-policy-compliance-checker/
    └── 005-inter-agency-knowledge-hub/

---

⚡ Quick Start (Any Accelerator)

# 1️⃣ Clone and navigate
cd newyork

# 2️⃣ Choose an accelerator
cd Constituent-Services-Agent  # or any other accelerator

# 3️⃣ Create virtual environment
python -m venv venv
venv\Scripts\activate  # Windows
# source venv/bin/activate  # Mac/Linux

# 4️⃣ Install dependencies
pip install -r requirements.txt

# 5️⃣ Run demo (mock mode - no Azure required)
python demo.py

# 6️⃣ Run web interface
python -m src.main

💡 Mock Mode: All accelerators work without Azure services using mock data for offline development. No API keys required to get started!

---

🔧 Key Technologies

Technology	Purpose	Badge
Azure AI Foundry	AI development platform
Foundry IQ	Intelligent document retrieval
Semantic Kernel	AI orchestration framework
Azure OpenAI GPT-4o	Language model for Q&A
Azure Document Intelligence	OCR and document parsing
Azure AI Search	Vector search with security filters
Microsoft Graph	Email and document access
Entra ID	Authentication and authorization
Flask	Web framework for APIs
Pydantic	Data validation

---

🧪 Testing & Evaluation

✅ Test Coverage

Accelerator	Tests	Status
Constituent Services Agent	43	✅ All Passing
Document Eligibility Agent	86	✅ All Passing
Emergency Response Agent	62	✅ All Passing
Policy Compliance Checker	14	✅ All Passing
Inter-Agency Knowledge Hub	38	✅ All Passing
Virtual Citizen Assistant (.NET)	22	✅ All Passing
Virtual Citizen Assistant (Python)	2	✅ All Passing
Total	267	✅ Production Ready

🤖 AI Evaluation Framework

• Quality Evaluators: Groundedness, Relevance, Coherence, Fluency
• Safety Evaluators: Content safety, PII detection
• Red Team Tests: Jailbreak, PII extraction, authority spoofing, hallucination

# Run tests for Python accelerators
cd [Accelerator-Directory]
python -m pytest tests/ -v

# Run tests for .NET accelerator
cd DotNet-Virtual-Citizen-Assistant
dotnet test

# Run AI evaluations
python -m shared.evaluation.eval_config

---

📈 Success Metrics

Accelerator	Key Metric	Target	Status
💬 Constituent Services	Response time	< 5 seconds	✅
💬 Constituent Services	Citation accuracy	> 95%	✅
📄 Document Eligibility	Processing time	< 2 minutes	✅
📄 Document Eligibility	Extraction accuracy	> 95%	✅
🚨 Emergency Response	Plan generation	< 5 seconds	✅
📋 Policy Compliance	Analysis time	< 30 seconds	✅
🔍 Knowledge Hub	Search response	< 3 seconds	✅

---

🎯 Hackathon Impact

👥 For Citizens

• ⚡ Faster answers: Get information about government services instantly
• 🌍 Accessible: Multi-language support, WCAG 2.1 AA compliant
• 📚 Transparent: See sources for all information provided

👔 For Agency Staff

• 📉 Reduced workload: AI handles routine inquiries, staff focus on complex cases
• ⏱️ Faster processing: Documents processed in minutes, not hours
• 🤝 Better coordination: Cross-agency visibility and emergency planning

🏛️ For Government

• ✅ Compliance: Built-in LOADinG Act and RAISE Act compliance
• 📈 Scalability: Handles high volumes during crises
• 📋 Accountability: Complete audit trails for all AI decisions

---

🤝 Collaboration & Access

Getting Access to This Repository

For Microsoft Enterprise Users: If you have a Microsoft enterprise account and are having trouble accessing this repository, please see our detailed Collaboration Guide for step-by-step instructions.

Quick Access Steps:

1. Ensure your GitHub account has 2FA enabled
2. Link your Microsoft enterprise email to your GitHub account
3. Request access from the repository owner (@msftsean)
4. For detailed instructions, see COLLABORATION.md

Contributing

We welcome contributions! Please see our Contributing Guidelines for:

• Code standards and best practices
• Pull request process
• Testing requirements
• Security considerations

Quick Start for Contributors:

# Fork and clone the repository
git clone https://github.com/msftsean/ai-hackathon-use-cases.git

# Create a feature branch
git checkout -b feature/your-feature-name

# Make changes and run tests
pytest tests/ -v  # Python projects
dotnet test       # .NET project

# Submit a pull request

---

📚 Additional Resources

• 🚀 Quick Start Guide
• 🤝 Collaboration Guide - For Microsoft enterprise users
• 📝 Contributing Guidelines
• 📋 Feature Specifications
• 🧪 Evaluation Framework
• 📖 Evaluation Guide
• 🔗 Azure AI Foundry Documentation
• 🔗 Semantic Kernel Documentation
• 🔗 Microsoft Accelerators

• 💡 Foundry-First Example Approach

---

🏛️ Shaping the Future of Responsible AI in New York State 🗽