I design and deliver end-to-end AI systems: from dataset construction and model training to production deployment.
I'm a data scientist and co-founder of DreamSpace Technologies — a software development agency in Batticaloa, Sri Lanka building AI and data systems for organisations that need them to work, not just exist. My work spans dataset construction, NLP models for native and English languages, computer vision, analytical dashboards, and end-to-end project delivery. I work directly with clients to take a problem from idea to deployed system.
Sri Lanka's most comprehensive digital hate speech dataset and detection system. Led the construction of a 55,000+ row multilingual corpus in Tamil, Sinhala, and Latin script across 12 fields. Fine-tuned open-weight LLMs for detection and counter-narrative generation in both languages. Managed two data collection teams of 20+ people.
An LLM-powered economic fact-checking assistant for journalists and the public. Verifies economic claims against official Sri Lankan government sources — Central Bank, Ministry of Finance — and translates complex policy into plain language across English, Sinhala, and Tamil. Built with an intelligent RAG pipeline featuring query expansion, relevance tuning, and reranking.
An ongoing white paper investigation into how popular LLMs exhibit bias when confronted with Sri Lanka's socially and historically contested topics. Uses structured prompting techniques, varied user roles, and multiple conversation languages to benchmark responses across models.
Developed a computer vision model to classify coffee bean sizes based on sub-centimetre differences. Also diagnosed a systemic hardware accuracy issue through diagnostic analysis — leading the company to overhaul their hardware setup and significantly improve overall system accuracy.