AI-Powered Digital Assistant for Monitoring & Evaluation Systems (UNICEF Context)
Kaloina Melodie Ravoahangilalao
Published
Complex monitoring systems are powerful — but difficult to navigate.
In the context of UNICEF WASH Madagascar, the MGMERL platform centralizes national data on water, sanitation, and hygiene. Yet many users struggle to access the right information quickly due to fragmented documentation and high system complexity.
I designed and built a Retrieval-Augmented Generation (RAG)–based intelligent assistant that simplifies access to monitoring tools without modifying the original system.
The solution relies on:
A Hive-based Data Lake (HDFS, Spark, MapReduce) to ingest heterogeneous data (JSON, PDF, PPTX, CSV)
n8n workflows for scheduled ETL automation with retry and error-handling mechanisms
Vector embeddings stored in Milvus for semantic search
FastAPI backend orchestrating RAG logic and authentication
React frontend with multilingual support and offline-first caching
Dockerized deployment with CI/CD via GitHub Actions
The assistant retrieves relevant documents through semantic search and generates contextualized answers in seconds, while preserving traceability and data integrity.
This project demonstrates full ownership: requirement analysis, architecture design, data engineering, AI integration, deployment, and documentation — delivered in a real humanitarian environment with infrastructure constraints.