Abdullah Tauqeer

I am an AI Researcher & Engineer specializing in the intersection of Computer Vision and Large Language Models (LLMs). My work ranges from building robust deep learning pipelines for medical imaging (tumor detection, semantic segmentation) to designing production-grade Retrieval-Augmented Generation (RAG) systems and Agentic workflows. I am passionate about tackling complex problems across multiple modalities, combining rigorous research evaluation with scalable, real-world deployment.

Email / Google Scholar / GitHub / LinkedIn

LLM/GenAI Focus

I build LLM systems that are trustworthy, scalable, and aligned with real user needs. My focus areas include:

RAG & Knowledge Systems: Designing retrieval pipelines, vector search, and document-grounded generation for accurate, up-to-date responses.
Agents & Workflow Automation: Creating tool-using agents that can plan, execute, and verify tasks across APIs, files, and databases.
Evaluation, Safety & Reliability: Building test harnesses, red-team suites, and monitoring to reduce hallucinations and measure quality over time.

Selected Projects

NLP/GenAI Projects

Zoom Intelligence Service [GitHub]

LangChain + LangGraph + FastAPI

Zoom transcription, summarization, and action-item extraction

Uses Zoom webhooks to trigger a FastAPI backend, queues processing with Celery + Redis, transcribes recordings via Whisper, summarizes with LangGraph, and stores results in Postgres for a Next.js dashboard.

opsrag-copilot [GitHub]

FastAPI + Next.js + Postgres + Celery

Enterprise RAG Copilot with human-in-the-loop agent actions

A production-ready RAG application featuring hybrid retrieval (BM25 + pgvector), role-based access control (RBAC), audit logging, and safe agent execution that requires human approval before tool actions. Includes built-in evaluation harnesses.

Computer Vision Publications and Projects

SERN-MIL: Selective Embedding Retrieval & Nuclei-Feature Aggregation based MIL for Prostate Gleason Grading [GitHub]

A. Tauqeer, N. Sheibani-Asl, A. Asif, A. Sadeghi-Naini

Submitted to Nature Methods, 2025

Proposed a novel Multiple Instance Learning (MIL) approach that leverages selective embedding retrieval to improve the accuracy and interpretability of prostate cancer grading.

Detection, Localization, and Staging of Breast Cancer Lymph Node Metastasis on Whole-Slide Images [GitHub]

A. Tauqeer, A. Asif, A. Sadeghi-Naini

Scientific Reports, accepted 2025

Developed a weakly supervised framework for precise metastasis detection in lymph nodes, achieving state-of-the-art performance on gigapixel pathology slides.

TexSegNet: A Hybrid Encoder and Decoder Model for Robust Nuclei Segmentation and Classification

A. Tauqeer, A. Asif, A. Sadeghi-Naini

EMBC 2025, accepted

Designed a hybrid encoder-decoder architecture for robust nuclei segmentation that generalizes well across varying stain protocols and scanner types.

Cell Tracking App [GitHub]

Python + Deep Learning

Live-Cell Time-Lapse Microscopy

An application designed for tracking live cells in time-lapse microscopy to analyze cell morphology and motility under varying conditions.

WSI Stitching App [GitHub]

Python + OpenCV

Digital Pathology Utility

Custom algorithmic tool for seamless image stitching of Hematoxylin and Eosin (H&E) stained images from in-lab microscopic scanners.

Skills

Large Language Models (LLMs) Retrieval-Augmented Generation (RAG) LangChain LangGraph LlamaIndex Agentic Workflows MCP Computer Vision Deep Learning Image Segmentation Generative Models Vector Databases LLM Evaluation Model Fine-Tuning Python PyTorch Docker CI/CD Pipelines

Experience

Research Engineer

Quantimb Lab

May 2025 - Present • Toronto, Ontario, Canada

Build performance-optimized computer vision pipelines for live-cell tracking in time-lapse microscopy, enabling quantitative analysis of cell morphology and motility relevant to cancer progression. Work with high-resolution imagery and large temporal datasets, focusing on efficient preprocessing, robust inference, and reproducible evaluation across staining protocols. Package cutting-edge workflows in Docker for consistent runs across environments, and optimize for real-time constraints using GPU-aware parallelism.

Graduate Researcher

Lassonde School of Engineering - York University

Sep 2023 - Present • Toronto, Ontario, Canada

Developed a novel deep learning framework for pathological nodal (pN) staging of breast lymph nodes, integrating nuclei features via LLMs, achieving a 4% improvement in F1 score. Designed a MaxViT-based H&E nuclei segmentation/classification network boosting F1-scores by 5.6%. Currently advancing prostate Gleason scoring using whole-needle biopsies with selective embedding filtering networks and attention-based MIL pooling. Proficient in PyTorch, solving bottlenecks on massive multi-resolution gigapixel WSIs.

Education

Master of Applied Science, Electrical & Computer Engineering Grade: A

York University

Sep 2023 - Jul 2025 • Toronto, ON

Co-supervised by Dr. Ali Sadeghi-Naini and Dr. Amir Asif.
Focus: Developed deep learning models for histopathology images to translate computational methods into actionable clinical and scientific insights.

Bachelor of Electrical Engineering CGPA: 3.7 / 4.0

National University of Sciences and Technology (NUST)

Sep 2019 - Jul 2023 • Islamabad, Pakistan