• Devised a unified A/B testing methodology and developed supporting dashboards to evaluate product features across the company's SaaS portfolio, reducing analysis turnaround time by 60%.
• Quantized Gemma 3 models for on-device inference on Android tablets, enabling secure semantic recommendations during product purchase flows on kiosk devices.
• Fine-tuned LLaMA-3 and DeepSeek variants using parameter-efficient fine-tuning (PEFT) and deployed inference endpoints on Oracle Cloud to support internal tools and 5 enterprise clients.

• Developed a transformer-based product recommendation model that increased conversion in the Just For You module by 2.3%.
• Built the AskDaraz conversational shopping experience with Azure OpenAI and scaled it to more than 200,000 queries per day.
• Engineered anomaly detection and root cause analysis pipelines for sales KPIs, saving $96,000 per year by reducing manual analysis.
• Developed dMart recommendation backend services and operator tooling, improving CTR by 30% over the hot-items baseline.

Scaled test-time compute of Falcon3 Mamba 7B using GRPO, improving GSM8K math performance by 37.2% over the base model and reaching more than 1,500 downloads in 2025.

• GRPO
• Reinforcement Learning
• PyTorch
• LLMs
• Fine Tuning

• Project Page

Built a mobile BCI application backed by an Azure-hosted deep neural network for classifying EEG signals into discrete actions; the project won Microsoft Imagine Cup 2022 in the EMEA Lifestyle category.

• Brain Computer Interface
• App Development
• Azure
• Deep Learning
• Feature Engineering

• Video Demo
• In the news

• GPA: 3.89/4.00

• GPA: 3.51/4.00