"Transform uncertainty into absolute confidence."
Advanced, governed, and traceable RAG (Retrieval-Augmented Generation) system for international trade. RAGulator is a 100% functional solution that unifies the Azure intelligence ecosystem to deliver grounded responses with immutable bibliographic citations.
intro_RaG_demo.mp4
RAGulator is not just a chat; it is a distributed orchestration of more than 13 Azure services working in harmony:
| Category | Azure Service | Strategic Role |
|---|---|---|
| Backend Compute | Azure App Service | Dynamic engine in .NET 10 with horizontal scaling. |
| Global Frontend | Azure Static Web Apps | Edge-optimized hosting for the React 19 SPA. |
| Database | Azure Cosmos DB (NoSQL) | Distributed storage for sessions, telemetry, and auditing. |
| Storage | Azure Blob Storage | Data Lake for source documents and file processing. |
| LLM Intelligence | Azure AI Foundry | Advanced reasoning engine using the GPT-5.4-mini model. |
| Data Ingestion | Azure AI Document Intelligence | Classification and extraction of tables/text from legal documents. |
| Vector Database | Azure AI Search | Semantic indexing and hybrid search (Keywords + Vectors). |
| AI Moderation | Azure AI Content Safety | Real-time shield against sensitive or unsafe content. |
| Identity & SSO | Microsoft Entra ID (Azure AD) | Corporate authentication and access control (RBAC). |
| Directory | Microsoft Graph API | Integration of user profiles and organizational structure. |
| Secrets | Azure Key Vault | Centralized management of production keys and certificates. |
| Observability | Azure Application Insights | Detailed performance telemetry and error traces. |
| Diagnostics | Azure Monitor | Infrastructure health analysis and critical alerts. |
| Performance | Azure Cache for Redis | Distributed cache layer (Cache-Aside) for instant dashboard loading. |
| Documentation | Scalar (OpenAPI) | Modern interactive API Reference integrated in the backend. |
Interactive API view (Scalar / OpenAPI)
The heart of the system uses the latest iteration of Azure efficient models, enabling complex reasoning with minimal response latency.
Uploaded documents are sent to Document Intelligence for high-fidelity OCR, chunked, and automatically vectorized so they are available in chat within milliseconds.
Every generated message includes a dynamically calculated Groundedness Score. All interactions are logged in Cosmos DB along with Application Insights logs, enabling full security audits.
The platform implements a persistent cache layer that stores telemetry and metrics results for 5 minutes, allowing the admin panel to respond in milliseconds after the first query.
- Strict CORS: Only the authorized Frontend on Static Web Apps can call the API.
- Native Entra ID: Secure login integrated with the organization's active directory.
- Content Safety: Automatic filtering of any response that does not comply with AI ethics policies.
- Zero-Secrets Policy: Passwordless authentication for Redis and Key Vault via Managed Identities.
RAGulator follows a clean microservices and layered architecture (Clean Architecture):
Frontend/: Single Page Application (SPA) built with React 19 and Vite 8. Implements a premium Design System with Glassmorphism and real-time data visualization via Recharts.Backend/: Robust Web API in .NET 10. Uses the official Azure SDK for native integration and exposes Technical Documentation (Scalar) for interactive endpoint testing.
The project uses GitHub Actions for a complete CI/CD pipeline:
- Build & Test: Compilation of the API and Frontend.
- Deploy: Automatic publishing to Azure App Service and Static Web Apps.
- Configuration: Dynamic secret management via Application Settings and protected environment variables.
- Control Dashboard: KPIs on latency, Groundedness Score, and processed documents.
- Dynamic Ingestion: Interface to upload PDFs that are automatically indexed in the vector engine.
- Security Auditing: Immutable logging in Cosmos DB of every interaction, including Content Safety alerts.
- Quality Management: Tracking of RAGAS metrics (Faithfulness, Relevance, Coherence, and Fluency).
Interface designed for a high-impact user experience:
- Deep Dark Theme: Professional aesthetic inspired by modern control centers.
- Micro-interactions: Smooth transitions and animated loading states for a fluid feel (60 FPS).
- Responsive: Fully adaptable to mobile devices for on-the-go queries.
This system represents the state of the art in Governed RAG Systems, demonstrating how the deep integration of Azure's native services creates a resilient, scalable, and above all, reliable platform for critical decision-making in international trade.
| Name | ||
|---|---|---|
| 👤 | Jimena Fioni |  |
| 👤 | Juan Pablo Urra Jara | |
| 👤 | Diego Alvarez | |
| 👤 | Fernando Pedernera | |