Project Consolidation and Interesting Finds

Over the past few weeks, the team has been focused on integrating and consolidating the various components of the Virtual Situation Room into a single, streamlined workflow.

Integration Highlights

  • Language Verification Agent Integration – We fully incorporated the agent into the main workflow, ensuring that content verification and AI-generated outputs maintain high quality across the system.

  • Frontend-Backend Connection – The dashboard frontend is now fully integrated with backend services, allowing seamless display of processed data and AI-generated reports.

  • Unified Pipeline Development – The team is consolidating modular agents and data components into a single pipeline, ensuring smooth operation from data ingestion, through processing, to final report generation and dashboard visualization.

Preparing for Client Handover

As the project approaches client delivery, the team was focused on making the system as user-friendly and reliable as possible:

  • Single-Click Workflow – Simplifying the pipeline so that data ingestion, processing, and reporting can all be triggered with a single click.

  • Chatbot Integration – The AI-powered chatbot is fully integrated, enabling users to interact and access the latest processed data seamlessly.

  • System Stability – Unit tests are being added across key modules to ensure robustness and maintainability prior to handover.


Guardrails prevent querying irrelevant content

Chatbot provides easily digestible insights on report data


Chatbot provides crisis analysis for quick overviews

Interesting Finds

  • AzureOpenAI Content Filtering – Some API calls initially failed due to content filtering policies. Adding notes about these policies within the Language Verification Agent’s system prompt resolved most issues. To make the system more robust, Azure's content-filtering guardrails were adjusted to ensure that sensitive language within humanitarian data was not incorrectly flagged by the LLM.

  • Model Routing – The team explored using Azure’s LLM model router, which automatically selects models based on query complexity—routing complex queries to a high-capacity model and simpler tasks to a smaller, more cost-efficient one. While this approach could optimize performance and cost, the team chose to assign specific models to each use case instead. This ensures greater control over token usage, answer depth, and maintains full transparency and traceability across the workflow.

The project continues to evolve into a fully integrated AI-powered dashboard for crisis analysis, giving the IFRC a Virtual Situation Room that combines automated data processing, real-time reporting, and intelligent insights.

Comments

Post a Comment

Popular posts from this blog

Finishing Touches and Deployment

Image
With the project deadline approaching, the team applied the finishing touches to the Virtual Situation Room, as well as ensuring robustness throughout the entire system. While some members were working on deploying the app onto Azure, the following features were being finalized: Resilience testing for the agent orchestration pipeline Admin panel for role-based access to key endpoints "What's New" feature that enables report comparisons Resilience Testing Resilience testing is implemented within the agent orchestration pipeline to ensure consistent and stable report generation under varying conditions. Given the inherent non-determinism of LLM-based systems, these tests focus on validating robustness rather than exact output matching. The resilience testing framework introduces controlled variations in input data, including formatting differences and reordered context. The pipeline is expected to produce semantically consistent reports despite these changes, ensuring that...
Read more

Expanding Data Integration and Backend Capabilities

Image
Development during this phase focused on strengthening the data infrastructure and backend capabilities that support the Virtual Situation Room.  A key milestone was implementing Azure PostgreSQL migration and integration into the core pipeline , allowing processed data and generated outputs to be stored and accessed more reliably across the system. In parallel, work began on building the backend application using Django , laying the foundation for a more robust application layer that connects the data pipelines, analytical services, and frontend interface. ReliefWeb Integration To expand the system’s information coverage, ReliefWeb was successfully integrated into the data ingestion agent, providing access to a broader and more consistent stream of crisis-related information. As a well-established humanitarian information platform, ReliefWeb offers structured and regularly updated reports from a wide range of trusted organisations, making it a valuable addition to the ingestion...
Read more