I work on AI systems that are both capable and legible, with reasoning you can actually inspect. I lead research engineering at Neuralgap, where my work spans molecular dynamics for viral capsid engineering, agentic explainability, and quantum approaches to molecular simulation.
My research sits at the intersection of computer vision, knowledge graphs and ontology-driven reasoning, agentic and multi-agent systems, computational biochemistry, and quantum computing. The connecting thread is explainability: my SLAAI 2025 best paper showed an ontology-backed vision pipeline matching VGG16-class accuracy on fruit disease detection while staying fully interpretable, with no black box required.
A hybrid "beyond black box" framework pairing a semantic knowledge base with a real-time vision pipeline. I designed a phytopathological ontology in OWL that formally codifies expert knowledge (disease symptoms, HSV spectral phenotypes, morphology) and reached 85.0% F1 on MangoFruitDDS, competitive with VGG16-class deep models while remaining fully interpretable.
A graph-based biomedical reasoning system that identifies and explains potential drug-drug interactions, combining LLM-driven natural language querying with semantic web ontologies for explainable insights. Deployed with LangGraph orchestration and inference tracking.
Fine-tuned pretrained ViT weights on a custom synthetic skincare dataset, reaching 92% test accuracy for oiliness classification from facial images, served through an interactive Streamlit app for real-time inference.
Fine-tuned Google Gemma-2B with LoRA and QLoRA for English-Sinhala translation, using PEFT to train inside a single-GPU memory budget. Instruction-tuned on a Sinhala-English dataset and released publicly on the Hub.
Demystifying Project AlphaFold - Where AI Meets Biology
What AlphaFold actually does, from the ground up: what
bioinformatics is, how the model gets from an amino acid sequence to a folded
structure, and what changed in AlphaFold 3.
Read on Medium ↗
Happy to talk about explainable AI, knowledge graphs, or anything at the boundary of machine learning and structural biology, especially if you are working on drug discovery.
