Overview

The InfoSec Copilot is a Microsoft Teams-integrated chatbot built to support internal users with security policy inquiries. Designed as a private, enterprise-grade assistant, it leverages Azure-native services for secure, scalable document retrieval and AI-powered responses.

The goal: eliminate friction in accessing InfoSec documentation, while ensuring the answers remain grounded in approved internal policies.

Architecture

Core components:

• Azure OpenAI (GPT-4): Handles natural language understanding and response generation.
• Azure Cognitive Search: Indexes security policy documents stored in Blob Storage for fast and relevant semantic retrieval.
• Azure Blob Storage: Stores the official InfoSec policy documents in markdown and PDF format.
• FastAPI: Powers the backend service to orchestrate document search, prompt construction, and LLM response delivery.
• Microsoft Teams Bot Framework: Provides the user interface within Teams for real-time question and answer interactions.

Optional integrations:

• Azure Monitor and App Insights for observability
• GitHub Actions for CI/CD and index automation

Use Cases

• “Can I use personal devices on company Wi-Fi?”
• “What’s our vendor onboarding security process?”
• “Do we use FedRAMP Moderate or High?”

These are resolved in seconds, rather than relying on lengthy documentation search or email threads.

Lessons Learned

• Semantic indexing in Cognitive Search dramatically improved relevance over traditional keyword-based search.
• Prompt tuning was key to grounding the model and avoiding hallucinations.
• Role-based access control ensured only appropriate documents and responses were available to users based on department and clearance.

Future Improvements

• Add user context awareness (e.g., role-based personalization)
• Version-aware policy responses
• Exportable query analytics for documentation improvement

Status

✅ MVP live and integrated with Teams
🔄 Ongoing improvements for RBAC, prompt tuning, and broader doc coverage