SLIIT COEAI Chatbot
An enterprise-level AI chat platform developed independently for research use at SLIIT, combining conversational AI, document intelligence, voice interaction, AI personas and image generation.
The problem
Students and researchers across several departments needed one capable AI workspace that could reason over their documents, maintain useful context and support more than text-only interaction.
The approach
I designed and built the complete platform around production-grade retrieval, interchangeable model providers, layered memory and real-time voice, while keeping prompts and assistant behavior manageable as the system evolved.
Built RAG pipelines with PostgreSQL, pgvector, embeddings and semantic document retrieval.
Implemented conversation summarization, semantic user memory and long-term contextual retrieval.
Integrated OpenAI, Anthropic Claude and xAI Grok behind a prompt-version management system.
Used the OpenAI Realtime API and WebSockets for live voice interaction.
Results
Actively used by more than 300 students across Psychology, Computing and Business.
Delivered document intelligence, AI personas, image generation and voice in one platform.
Created a production-oriented multi-provider system instead of a single-model prototype.
Explore more AI systems and product engineering case studies.