Danuka Wickramasinghe

Built a modular LLM agent on GCP with LangChain, Langraph, and Gemini, significantly reducing task times and subcontractor dependence. Improved RAG response quality by integrating multimodal data and applying advanced RAG techniques, achieving over 91% faithfulness and 90% context precision.

Presented machine learning findings to non-technical stakeholders and developed solutions to meet their requirements. Developed ML/DL models for the pricing team, ensuring FCA compliance for risk evaluation and pricing refinement.

Delivered end-to-end computer vision solutions for fishery measurements and deployed them as web applications. Built sentiment-analysis and NLP models using Transformers and LSTMs, and forecasted electricity demand by combining multiple ML algorithms.

Activities and societies: Modules: Statistical Learning Methods, Systems Engineering, Intelligent Cyber-Physical Systems, Search and Optimization, Logic and Automated Reasoning, Data Analytics and Visualization, Ethical, Regulatory, and Social Aspects of AI, Deep Learning.

Developed a scalable Machine Learning solution using autoencoders and pose estimation, deployed via AWS to identify alcohol-intoxicated airport passengers with 85% accuracy. Created a web app dashboard for real-time result display. Individual Research Project: Cognitive radar - Dynamic optimization of radar performance based on the perception-action cycle of reinforcement agents.

Activities and societies: Modules: Aircraft Systems, Air Transport Economics, Project Engineering and Management, Analytical Techniques, Electronics and Control, Aerospace Design Methods and Materials, Engineering and Mechanical Principles.

Designed an optimal control system for a quadcopter using MATLAB and Simulink. Selected the best-suited aerofoil for a glider, developed a mathematical model to calculate the gliding range, and collaborated with team members to validate numerical solutions.