My research focus bridges the gap between theoretical AI and industrial-scale deployment. With over 20 years of experience, my work has evolved from low-level systems optimization to architecting Agentic AI and Conversational Systems that drive enterprise value. I specialize in building resilient AI frameworks that operate reliably at national scales and within high-stakes environments like Healthcare and Telecommunications.
Currently, I lead R&D initiatives centered on Agentic AI workflows. This involves designing multi-agent orchestration frameworks that utilize Large Language Models (LLMs) for complex decision-making, autonomous reasoning, and enterprise process automation. My focus is on the reliability, observability, and scalability of these agents in production environments.
I have pioneered the development of end-to-end NLP and Speech AI pipelines. A hallmark of this work was the national-scale deployment of Urdu ASR and NLU models, automating analytics for 96,000 daily interactions with 85% accuracy. This research addresses the challenges of resource-constrained languages and real-time semantic analysis in contact center environments.
Mirza O. Beg et al. Advanced Multimedia Analytics for Enterprise Decision Support. Multimedia Tools and Applications. 2025.
Mirza O. Beg et al. Resource-Efficient NLU for Low-Resource Linguistic Ecosystems. Language Resources & Evaluation. 2023.
Saira Qamar, Hasan Mujtaba, Mirza O. Beg. Relationship Identification between Conversational Agents using Emotion Analysis. Cognitive Computation. 2021.
Mubashar Nazr Awan, Mirza O. Beg. TOP-Rank: A Topical PositionRank for Extraction and Classification of Keyphrases. Computer Speech and Language. 2020.
In the domain of Oncology and Clinical AI, my research focuses on transparency and interpretability. By engineering explainable NLP models, we provide clinicians with actionable insights grounded in medical evidence, fostering trust in AI-driven decision support systems for high-stakes healthcare outcomes.
Building on my foundational work in compiler theory, my research in Computational Sustainability explores the intersection of Machine Learning and energy efficiency. We develop tools to help developers minimize the carbon footprint of AI models and software applications on power-constrained mobile and IoT devices.
Ahmed Uzair, Mirza O. Beg, et al. WEEC: Web Energy Efficient Computing: A Machine Learning Approach. Sustainable Computing: Informatics and Systems. 2019.
Hareem Sahar, Mirza O. Beg. Towards Energy Aware Object-Oriented Development of Android Applications. Sustainable Computing: Informatics and Systems. 2019.
My earlier research established the technical bedrock for modern AI infrastructure through combinatorial optimization, graph theory, and network protocol formalization.
Focused on automatic parallelization for clustered architectures and cache-conscious data placement. This work utilized constraint programming to solve temporal and spatial scheduling problems in integrated manners.
Mirza Beg, Peter van Beek. A Graph Theoretic Approach to Cache-Conscious Data Placement. ISMM '10.
Mirza Beg. Instruction Scheduling on Multicores. PLDI '10 (ACM SRC Winner).
Collaborated on the Axiomatic Basis for Communication, formally modeling communication paradigms to enable correctness proofs and formal analysis of global-scale protocols.
Martin Karsten, S. Keshav, Mirza Beg, et al. An Axiomatic Basis for Communication. SIGCOMM '07.
Mirza Beg. FLECS: A Framework for Rapidly Implementing Forwarding Protocols. COMPLEX 2009.