Final Year Project

Project Title

Alkhidmat Public Chat Portal

Description

The AI-powered multilingual chat portal for Alkhidmat Foundation addresses manual query-handling delays by providing a centralized, inclusive solution. Using a Self-RAG (Retrieval-Augmented Generation) and Agentic AI pipeline, it automates responses for donor, healthcare, and general domains in English, Urdu, and Roman Urdu. The system utilizes multilingual-e5-base embeddings and pgvector for semantic search, with OpenAI and Alif for generation. To ensure accuracy, it employs a domain classification and confidence scoring engine that fuses retrieval quality with token probability. Key benefits include 24/7 accessibility for underserved communities, reduced staff workload, and automatic human-agent escalation for complex queries. Moreover, there is a dedicated admin dashboard to view LLM analytics and update the RAG Knowledge Base.