COPAI knowledge assistant

Overview

The Community of Practice on Artificial Intelligence (COPAI) is a multi-stakeholder AI research and innovation platform in Cambodia, established by UN ESCAP in collaboration with the Ministry of Industry, Science, Technology & Innovation (MISTI). Within its AI Innovation Lab, I served as Lead AI Research and Innovation across Gen I and Gen II in a volunteer, part-time, hybrid capacity. I led AI system development and built the locally hosted assistant to connect Cambodian AI and policy documents.

My Contribution

• Led eight team members across COPAI Gen I and Gen II, supervised by Dr. Chen Sovann (MISTI), H.E. Dr. Seingheng Hul (MISTI), and Mr. Rafael Torquato Cruz (UN ESCAP); led AI system development, built the assistant on MISTI infrastructure, and led team data collection and knowledge-base preparation.
• Connected document processing, embeddings, vector retrieval, reranking, and local language-model inference.
• Worked with application access controls and monitoring on MISTI GPU infrastructure.

Key Features

• Document ingestion and preparation
• Vector retrieval and reranking
• Source-grounded chat
• Role and group access controls

Technical Implementation

• Open WebUI for chat and retrieval orchestration
• BGE-M3 embeddings, Qdrant vector storage, and BGE reranking
• vLLM inference on NVIDIA A40 infrastructure
• Docker services; architecture includes PostgreSQL, MinIO, Redis, and Langfuse with Prometheus/Grafana monitoring

Results

Limitations

• No independently verified retrieval benchmark, adoption count, or production uptime is claimed.
• Initial Khmer support was delivered but its answer quality was not formally evaluated; broader integrations remained roadmap directions.
• Source citations and local hosting alone do not establish factual reliability or fairness.