Vihanga G.

I'm Vihanga, an AI engineer and architect with experience building systems and solutions applying LLMs, RAG, and agentic and multi-agent approaches. I began building my expertise in the field working with Machine Learning over a decade ago, followed by my work as a researcher in Deep Reinforcement Learning. In 2021 I began freelancing, carrying out development and consulting work, expanding my expertise as I delivered for clients across different fields. Most of my recent work has been in RAG and agentic systems: retrieval pipelines with domain-specific knowledge integration, multi-agent orchestration for AI platforms, knowledge graph construction with quality benchmarking, and the evaluation and scoring layers that sit around all of it. I've also built semantic clustering that handles 50K+ records with sub-second inference, RL agents and LLM content generation for a real-time virtual world platform demoed at CES, and on-device ML pipelines optimised for Apple Silicon and CoreML. I work across research and development, and that blend tracks back to my history. Following an undergrad in engineering, I moved to the field of AI in 2015 with an MSc in Advanced Computer Science specialising in Artificial Intelligence at the University of Manchester, and went from there into PhD research applying deep reinforcement learning to realistic virtual character animation. The culmination of my research was RLAnimate, a first of its kind RL framework for training agents that animate lifelike virtual humans in real time, along with peer-reviewed publications and a best paper award. My research was always about building working systems, and the freelance and consulting work since has tracked the shift in the field: from classical deep learning and RL through to language model driven systems, retrieval, and agents. Across these engagements, the work has also covered: * Structuring AI agents for real-time responsiveness and role-based behaviour * Designing embedding and retrieval systems for context-aware AI * Evaluation and scoring systems for quality-gated generation * Cloud deployment and infrastructure: containerised backends, serverless, GPU and CPU-based hosting * On-device and local inference: Apple Silicon, CoreML, CPU-optimised pipelines