Vihanga – LLM, AI agent development, Machine learning

A next-generation intelligent assistant platform: a companion app whose intelligence runs fully on-device on iPhone, so user data never leaves the phone. The product combines LLM reasoning, knowledge-graph-based retrieval, semantic clustering, and memory systems under strict privacy requirements and a regulated healthcare pathway. I was the AI architect for the platform and the sole developer of its intelligence layer.

• Led applied AI architecture for a next-generation intelligent assistant platform, integrating large language models (LLMs), knowledge representation, and memory systems.
• Designed and implemented components across the AI stack, including natural language understanding, semantic clustering, retrieval-augmented generation (RAG), and adaptive behavior modeling.
• Delivered CoreML-compatible machine learning pipelines and optimized on-device inference using Apple Foundation Models and MLX.
• Built the evaluation and benchmarking layers around the AI stack, so every component had measurable behavior before release.

A real-time 3D virtual world platform where AI-driven characters interact live with users: contextual dialogue grounded in domain knowledge, role-based character behavior, and procedural world building. The platform's AI systems were demonstrated at CES 2024 and at an NFL Super Bowl-affiliated Gamerfest event as part of successful investor and public showcases.

• Designed and implemented reinforcement learning agents and LLM-driven content generation systems for the GURU virtual world platform, handling real-time interaction and procedural world-building across complex 3D environments.
• Built retrieval-augmented generation (RAG) pipelines integrating domain-specific knowledge bases into interactive AI characters, enabling contextual dialogue and dynamic behavior across multiple virtual scenarios.
• Developed prototype AI systems demonstrated at CES 2024 and NFL SuperBowl-affiliated Gamerfest events, contributing directly to successful investor and public showcases across two major industry exhibitions.

A commercial mobile app on the App Store with personalized wine recommendations at the core of the product. The recommendation engine serves personalized results to end users at scale, built from sparse user signals and multi-source catalog data with inconsistent schemas. I built the engine end-to-end, from the first proof of concept through to the production API serving the app.

• Led development of an LLM-augmented deep learning recommendation engine, building the full pipeline from multi-source data aggregation and feature engineering through to model training and evaluation.
• Deployed the recommendation system as a production API integrated into a commercial mobile application, delivering personalized wine recommendations to end users at scale.
• Ran the system in production for over three years, tuning against live results, with distribution monitoring that caught a data-quality incident within a week for a clean rollback.

A reinforcement learning framework for training agents that animate lifelike virtual human characters in real time, developed as my PhD research. It is designed for game developers and researchers working in agent-based control and data-driven animation, and the project page carries renders, papers, and the technical story.

• Developed a novel model-based reinforcement learning approach using learned latent dynamics models, significantly reducing training time and motion capture data requirements compared to state-of-the-art methods.
• Built agents that operate frame by frame to portray varied social behaviors and conversational gestures, enabling live animation synthesis from arbitrary voice input.
• Published peer-reviewed work at venues including ICANN.