At Asrix BHD, I performed finetuning on llm’s for specialized internal tasks along with building and implementing agent architectures to automate streamline staff tasks.
I developed and optimized a computer vision pipeline for pedestrian and transport detection using YOLO architectures, tuning inference for real-time webcam feeds. I also fine-tuned a local Mistral 7B model on HuggingFace datasets to summarize, classify, and extract structured telemetry from unstructured logs.
On VoltStock, I built an inventory platform with a natural-language AI co-pilot for warehouse staff. I designed an MCP server to connect language models to MongoDB and integrated the Agent Payments Protocol for autonomous procurement.
For my PCB Defect Detector & Explainer capstone, I combined a custom-trained YOLOv8 model with vision language models to detect and explain manufacturing defects. I also built a dashboard with live camera streaming, scan history, and automated analytics reports.
I developed UnderWrite.AI, a document-analysis application for business loan and insurance risk assessment, using a ReAct agent and a retrieval-augmented knowledge base. Alongside my AI engineering work, I’ve taught coding at GotGame (WeCode Academy, client: Nokia) and am completing a BSc Computer Science (Hons) specializing in Artificial Intelligence.
United Kingdom
Malaysia