Master of Science: Computer Science

2022 - 2023

University of Alberta, Edmonton, Canada

Master's Thesis Title: Locating anomalies in aerial multi-spectral imagery

Bachelor of Technology: Computer Engineering

2017 - 2021

Aligarh Muslim University, Aligarh, India

Bachelor's Thesis Title: Nowcasting of Multispectral Satellite Imagery from INSAT-3D using Neural Networks

Alberta Machine Intelligence Institute (AMII) | Machine Learning Resident

October 2024 – Present

• Engineered an end-to-end data processing pipeline to standardize and unify large-scale (5,000+ hours) heterogeneous audio datasets from diverse sources, paving the way for fine-tuning Large Acoustic Models (LAMs) for voice-based diagnostics and multi-disease detection.
• Developed an intelligent audio pre-processing system integrating Whisper-based speech recognition, speaker diarization, and Phi-4-based transcript analysis to automatically extract relevant patient speech segments from clinical conversations, preparing high-quality input for diagnostic model training.
• Designed a hybrid Large Acoustic Model (LAM) architecture that integrates audio transformer embeddings, OpenSMILE-based voice metrics, and ModernBERT-encoded demographic features via self-attention; leveraged contrastive learning to align the fused audio features with text-encoded patient disease descriptions, achieving an F1-score of 80% and an MCC of 70%.

Vector Institute | Machine Learning Associate - Generative AI

May 2024 – September 2024

• Spearheaded the development and deployment of a personalized medical scribe using Llama-3 architecture, utilizing Vertex AI on GCP for efficient data processing and vLLM for optimized model serving and low-latency inference.
• Implemented distributed training using Fully Sharded Data Parallel (FSDP) for fine-tuning and preference tuning of the Mistral-7B and LLaMA-3 models, 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 few-shot performance benchmarks.

Aerium Analytics | Machine Learning Developer

May 2023 – April 2024

• Developed multi-spectral FOD detection for drone imagery using normalizing flows and masked image modeling for robust feature extraction.
• Achieved 99.28% IoU with Segformer for precise grass segmentation on airport runways, enhancing overall system precision.
• Optimized a YOLO-NAS model via TensorRT conversion, boosting inference speed by 15% and deploying efficiently with DeepStream on Jetson Orin Nano.
• Deployed models as versioned FastAPI endpoints in Docker within an Agile framework, ensuring seamless integration and maintenance.

University of Alberta | Graduate Research Assistant

October 2022 – April 2023

• Implemented Implicit Neural Representation techniques to enhance spectral data richness in multi-spectral images
• Created algorithms for transforming multi-spectral data from Sentinel-2 and Landsat into hyper-spectral images
• Achieved 420% improvement by increasing spectral bands from 5 to 26

Fatima Fellowship | Natural Language Processing Fellow

June 2022 – February 2023

• Developed the PyDebiaser library for addressing bias in transformer models (GPT2, BERT, GPT-Neo)
• Demonstrated impact of bias mitigation techniques on large language models
• Fostered ethical AI practices and fairness in NLP applications

Indian Institute of Technology, Delhi | Machine Learning Engineer

September 2021 – March 2022

• Developed Intelligent Edge Device prototype for visually impaired pedestrians
• Ported PyTorch models to ONNX, reducing model size by 50%
• Achieved near real-time detection on Raspberry Pi with Intel Neural Compute Stick

Indian Institute of Information Technology, Allahabad | Research Intern

Dec 2020 – Feb 2021

• Worked on the development of Explainable AI model for Seizure Detection using EEG data.
• Reviewed literature on Explainable AI techniques.
• Designed Residual Depthwise Separable convolution-based architecture for seizure detection.

King Fahd University of Petroleum & Minerals, Dhahran | Research Intern

Jun 2020 – Nov 2020

• Worked on File Fragment Classification which is part of digital forensics and data carving.
• Developed a Depthwise Separable Convolution based model for efficient classification of data fragments, the model is 24 times faster than the current state-of-the-art.

Interdisciplinary Biotechnology Unit, AMU | Undergrad Researcher

Dec 2019 – March 2020

• Worked on the localisation of intrinsically disordered regions in protein sequence given in Fasta format using residual ConvLSTM network.
• Reviewed literature on deep learning techniques for drug repurposing specifically for Covid-19.