Resume

Senior Data Scientist

Capital One

• Developing a generalized evaluation framework to benchmark LLM and agentic system performance.
• Improve validation components for a production multi-agent system, verifying factual accuracy of generated outputs in high-volume customer-service workflows.

AI Research Intern

Comcast

• Built distributed data pipeline using PySpark to process large-scale conversation datasets and added LLM supervision into clustering frameworks using LangChain, LLM Tool/Function Calling to improve the categorization.
• Designed a hierarchical, multi-layered clustering framework by integrating LLM-based verification and summarization into agglomerative clustering to create descriptive cluster hierarchies.
• Developed LLM-based evaluation frameworks for comparing the different unsupervised clustering methods.

Graduate Researcher

Virginia Polytechnic Institute and State University

• Creating a large-scale multimodal (text-video-audio) evaluation dataset & LLM-based method for event/entity extraction and relation extraction (event matching/search) to reduce the information gap between modalities.
• Improving the object and part detection of LLMs using LLM post-training, i.e., reinforcement learning (PPO, GRPO).
• Uncovering security vulnerabilities in multimodal LLMs via novel adversarial attack in realistic black-box setting.

Machine Learning Engineer

IQVIA

• Created IQVIA's AI-Assistant by fine-tuning (SFT) LLaMA model on business-specific use cases to enhance user experience and product engagement.
• Finetuned LLM via QLoRA and quantized with AutoAWQ that reduced response time by 54% and memory usage by 73%.
• Designed and deployed production LLM serving infrastructure using vLLM on AWS cloud, implementing autoscaling, monitoring, and GPU inference optimization.
• Experimented retrieval augmented generation (RAG) based solution to accommodate fast changing information.
• Enhanced IQVIA's recommendation algorithms suite, generating over $10M in annual revenue, and contributed to the deployment & MLOps pipeline of these models using Kubeflow & Kubernetes.

Research Intern

Xu Lab - Carnegie Mellon University

• Improved the understanding of different sentence components in video-language joint learning tasks
• Devised a solution to detect anomalous events in automated tools during experiments in a laboratory using recorded videos.

IEEE Video and Image Proceesing Cup 2022

IEEE Signal Processing Society

Synthetic Image Detection

• First Runner up

HP Presents Kaggle Days World Championship 2022

Kaggle

• Winner of Shanghai Regional
• Selected as one of the top 36 data science teams globally to participate in the final event at Barcelona, Spain

IEEE Signal Proceesing Cup 2022

IEEE Signal Processing Society

Differentiation between original human speech and synthetic speech.

• Winner
• Presented our solution at ICASSP 2022, Singapore

Deep Chimpact

DrivenData

Depth Estimation of animals from Images for Wildlife Conservation

• Winner
• 1st Place out of 313 teams

NFL 1st and Future - Impact Detection

Kaggle

Detection of helmet impacts between players in videos of NFL games.

• 25th Place out of 460 teams
• Silver Medal Winner at Kaggle

SIIM-FISABIO-RSNA COVID-19 Detection

Kaggle

Identify and localize COVID-19 abnormalities on chest radiograph images

• 4th Place out of 1305 teams.
• Gold Medal Winner at Kaggle
• Best Student Team Award Winner

Master's of Science

Virginia Polytechnic Institute and State University

Computer Science

CGPA: 4.00/4.00

• Information Extraction and Retrieval
• LLM Security
• LLM Training

Bachelor of Science

Bangladesh University of Engineering and Technology

Computer Science and Engineering

• Data Structures and Algorithms
• Software Development
• Machine Learning