Kunwar Muhammed Saaim Email & Phone Number

Edmonton, Alberta, Canada

Toronto, Ontario, Canada

• Spearheaded the development and deployment of a personalized medical scribe using Llama-3 architecture, utilizing Vertex Al on GCP for efficient data processing and vLLM for optimized model serving and low-latency.
• Implemented distributed training using Fully Sharded Data Parallel (FSDP) for fine-tuning and preference tuning the Mistral-7B and LLAMA-3 models, while benchmarking their efficacy across multiple LLMs.
• Designed and refined instruction prompts for MedLM (PaLM 2) and Claude Sonnet-3.5, implementing recursive refinement and chain-of-thought methodologies to achieve state-of-the-art zero-shot and one-shot performance.

Calgary, Alberta, Canada

• Developed Foreign Object Debris (FOD) detection solutions for multi-spectral drone images, utilizing normalizing flows to model FOD as anomaly
• Implemented masked image modeling techniques to extract essential features, resulting in a marked improvement in the system’s precision and versatility, contributing to more effective FOD detection
• Achieved a remarkable 99.28% Intersection over Union (IoU) with the Segformer model for precise grass segmentation in airport runway images
• Enhanced the reliability and flexibility of the machine learning model by implementing versioning capabilities within FastAPI endpoints, ensuring smooth deployment and maintenance processes.
• Collaborated within an Agile framework to swiftly deploy and integrate models as FastAPI endpoints, ensuring seamless operational integration within Docker container applications
• Optimized and converted a custom-trained YOLO-NAS model to TensorRT, achieving a 15% increase in inference speed.

Edmonton, Alberta, Canada

• Conducted research on spectral super-resolution of multi-spectral satellite images using Implicit Neural Representation (INR) techniques.
• Developed advanced methodologies and algorithms for achieving hyper-spectral images from Sentinel-2 and Landsat multi-spectral data, contributing to advancements in remote sensing and image processing.

United States

• Played a pivotal role in measuring and mitigating bias in transformer models, including GPT2, BERT, and GPT-Neo, and actively contributed to the development of PyDebiaser library.
• Showcased a profound comprehension of how bias mitigation techniques influence the language modeling scores of large language models (LLMs), emphasizing the importance of responsible AI in natural language processing.

New Delhi, Delhi, India

• Actively contributed to the development of an Intelligent Edge Device prototype aimed at empowering visually impaired pedestrians with enhanced navigation, safety, and social inclusion, successfully implementing.
• Demonstrated adaptability by seamlessly porting and quantizing PyTorch models to ONNX, and subsequently integrating refined models into a multi-threaded control system, resulting in the optimization of the device’s.

• Led the development of an Explainable Deep Learning model for Seizure detection, delivering a groundbreaking solution that increased detection accuracy by 15% compared to existing models.
• Conducted a comprehensive literature review on Explainable AI, equipping the team with valuable insights and best practices in this emerging field.
• Pioneered the creation of an Explainable Depthwise Separable Residual Convolutional Neural Network for Seizure detection, achieving a 20% reduction in false positives while maintaining high sensitivity levels.

Dhahran, Saudi Arabia

• Designed and executed experiments on file fragment classification.
• Developed a Depthwise Separable Convolution based architecture for efficient classification of data fragments.
• Achieved 24 times improvement in inference time compared to current state-of-the-art while maintaining better accuracy.

Aligarh

• Used hierarchical approach to predict intrinsically disordered proteins using Attention ConvLSTM.
• Examined the clustered embedding space to validate the learned dense representations with in-vitro experiments.
• Reviewed literature on deep learning techniques for drug repurposing specifically for Covid-19.

Master Of Science, Computing Science

Bachelor Of Technology, Computer Engineering

Senior Secondary School

Secondary School