Architecting the future of AI through production-grade NLP, LLMs, and CV Systems. Bridging the gap between cutting-edge research and scalable enterprise solutions.
I am an AI/ML Engineer driven by the complexity of human cognition and its digital replication. My work explores the intersection of Empathetic AI and high-performance system architecture.
Specializing in multi-task learning and RAG-based systems, I build production backends that don't just process data—they understand context. From optimizing CUDA kernels to fine-tuning LLMs, I focus on the full stack of intelligence.
Uva Wellassa Univ / B.ICT (Hons)
Scaling inference for 5,000+ daily ops.
Optimized through efficient query indexing and caching.
A framework for emotion intensity scoring.
A detailed trace of specialized engineering and AI development initiatives.
Engineered multi-agent GenAI pipelines for bulk content generation, reducing manual entry time by ~70% and processing 10,000 assets weekly.
Developed multiple ML-powered Chrome extensions to automate dynamic web scraping, improving extraction reliability by 90% over legacy scripts.
Architected a real-time multi-vendor platform with secure payments and WebSockets, supporting 1000 concurrent users with sub-second latency.
Fine-tuned YOLO architectures (v5) for custom object detection, increasing Mean Average Precision (mAP) by 20-30% through rigorous hyperparameter optimization.
Conducted Exploratory Data Analysis (EDA) on 1,000+ images to identify class imbalances,implementing augmentation strategies that reduced edge-case prediction failures by 30%.
Collaborated on deployments, reducing inference time by 30% through optimization techniques.
Developed and optimized REST APIs for a microservices e-commerce platform, reducing average endpoint response time by ~30% through efficient database query indexing.
Diagnosed and resolved critical data flow bottlenecks between services, eliminating cross-service timeouts and improving data synchronization accuracy to 99%
Enhanced system reliability under high load (5,000+ daily requests) by implementing Redis caching, which reduced core database load by approximately 40%.
Deploying state-of-the-art architectures to solve real-world complexities. Engineering for precision, latency, and scale.
Emotion & Sentiment Analysis system with shared-encoder MTL architecture. Reduced memory footprint by 60% and boosted F1-score by 15% via progressive unfreezing and dynamic loss weighting.
Extractive text summarization engine with custom TextRank algorithm. Achieved <50ms latency via NumPy optimizations and 80% information retention. Fully local FastAPI inference service.
Cross-platform GPU computing library built in Rust. CUDA-free GPU tensor runtime using wgpu. Accelerated Python tensor ops by ~50× via PyO3 GPU bindings with 40% reduced compute overhead.
AI-powered document application for intelligent note-taking and summarization. Integrates LLM capabilities for content generation and organization.
Production RAG system using LangChain/LangGraph. Improved retrieval precision by 40% and reduced LLM hallucinations by 30% through robust document parsing and chunking strategies.
Fine-tuned YOLOv5 for custom object detection. Increased mAP by 20-30% through hyperparameter optimization and augmentation strategies on 1,000+ images.
A multidimensional stack optimized for performance, scalability, and robust inference.
Exploring the frontiers of conversational intelligence and data-driven insights.
Uva Wellassa University of Sri Lanka
Research on building emotion-aware conversational AI systems with emotion intensity scoring and optimized empathetic responses.
Undergraduate Research Symposium of Technology & 3rd Applied Sciences (URSTech & APSURS 2024)
9th International Conference of Sabaragamuwa University of Sri Lanka, Colombo
Available for high-impact AI/ML engineering roles and research collaborations.