Dr. Shahnawaz Ahmed

CV of Shahnawaz Ahmed, Senior Deep Learning Researcher at Embedl. Neural-network inference optimization for edge hardware: quantization, pruning, and real-time deployment of VLM/VLA models on NVIDIA, AMD, and Qualcomm platforms. PhD from Chalmers University of Technology.

Dr. Shahnawaz Ahmed

Email: shahnawaz.ahmed95@gmail.com | Google scholar | GitHub:quantshah | LinkedIn: quantshah

Deep learning researcher (PhD) working on real time AI model inference on edge hardware.

I currently lead the development of Embedl Deploy, a tool that solves deployment and quantization of PyTorch models on edge hardware. I am also working within Embedl’s research team optimizing LLMs, VLMs and VLAs such as (Physical Intelligence π-0.5) for real-time robot control, by developing novel quantization and optimization methods. My experience spans open-source software and peer-reviewed research, working mainly in PyTorch, JAX, and TensorRT, with hands-on inference experience on hardware such as NVIDIA AGX Orin, AMD Strix Halo, Qualcomm Snapdragon NPUs, and TI TDA4 chips.

EDUCATION

EXPERIENCE

Embedl, Sweden | Oct 2023 – Present

Senior Deep Learning Researcher | Product owner (previously Deep Learning Research Engineer)

  • Optimize inference of vision-language models (VLMs) and vision-language-action models (VLAs), e.g., Physical Intelligence π-0.5: profile and eliminate performance bottlenecks to bring robot control policies to real-time execution (10 Hz) on multiple hardware platforms, e.g., NVIDIA Thor and AMD Strix Halo GPUs.
  • Full-integer (INT8) quantization of VLMs such as Qwen 2.5 VL 7B while preserving accuracy, to run on NPUs.
  • Research and develop deep learning model optimization techniques: quantization (PTQ and QAT), pruning, and hardware-aware neural architecture search. Co-authored FlashHead, an efficient drop-in replacement for the classification head that speeds up language model inference.
  • Product owner of embedl-deploy in a team of six. Embedl Deploy is a Python toolkit that enforces hardware and compiler constraints directly in PyTorch (hardware-accurate IR, PTQ/QAT quantization) so models deploy on edge devices with predictable performance and accuracy. Drive design decisions for the tool and build compelling demos showcasing its value, e.g., deployment of SAM 3 and SAM 3D models.
  • Build relationships with chip vendors such as Axelera, Lattice, and AMD to enable fast deployment on their hardware. Drive partnerships and develop backends for embedl-deploy that simplify deployment of AI models for different hardware targets.
  • Directly responsible for the success of customer projects, e.g., liaising with SAAB, Kodiak and BMW working closely with their engineering teams on problems ranging from deploying vision and perception models on NVIDIA Orin GPUs, to language models on Samsung and Qualcomm NPUs.

Xanadu Quantum Technologies, Canada | May 2023 – Present

Research collaborator within Quantum Machine Learning

  • Exploring the feasibility of quantum machine learning (QML) to solve real-world problems.
  • Co-authored a benchmarking paper evaluating 9 quantum models with 200,000 executions on an HPC cluster, prompting critical discourse in QML research. Blog
  • Co-authored a paper on developing a method to train quantum generative models scaled upto 1000 qubits link.
  • Collaborated on scaling quantum models for HPC implementation with National Energy Research Scientific Computing (NERSC).

Chalmers University of Technology, Sweden | Oct 2018 – Sep 2023

Ph.D. – Quantum Information and Computing

  • Developed machine learning and optimization techniques for quantum computing applications: Generative neural networks for quantum tomography PRL 127, 140502 (2021), a Riemannian optimization algorithm for learning quantum processes from data PRL 130, 150402 (2023). Also collaborated on various experiments such as continuous-variable quantum gate optimization to create non-classical photonic states (GKP, CAT) PRX Quantum 3, 030301 (2022), performance analysis of superconducting quantum gates npj Quantum Inf 9, 44 (2023).

  • Developed a Bayesian data analysis technique in collaboration with Prof. Ann L’Huillier’s (Physics Nobel laureate 2023) team at Lund University to analyze experimental data in the first ever quantum state tomography of photoelectrons.

Xanadu Quantum Technologies, Canada | May 2022 – Aug 2022

Quantum Algorithms Intern (Funded by Mitacs Accelerate Fellowship)

  • Developed an implicit differentiation approach for variational quantum algorithms, applied to quantum chemistry and condensed-matter systems (arXiv:2211.13765).
  • Contributed to the PennyLane software and developed tutorials explaining quantum machine learning techniques (e.g., a data re-uploading classifier).

Quantum Toolbox in Python (QuTiP) | 2016 – Present

Core Contributor and Admin Team Member

  • Maintained and enhanced QuTiP, a widely-used quantum simulation library.
  • Developed JAX-based tools for quantum physics applications (qgrad, qutip-jax).
  • Contributed to the development of software for pulse-level simulation of quantum circuits (Quantum 6, 630 (2022)).

Cluster for Pioneering Research, Riken, Japan | Jul 2017 – Jul 2018

Theoretical Quantum Physics Intern

  • Developed neural network-based constrained optimization techniques (e.g., Sudoku solving).
  • Contributed to developing code and software to simulate large open quantum systems, contributing to publications (Phys. Rev. A 98, 063815).

Google Summer of Code (Python Software Foundation) | May 2016 – Aug 2016

Intern – DIPY Project

  • Developed a Python module for MRI reconstruction (demo).
  • Mentored subsequent GSoC projects for QuTiP, contributing to publications (Quantum 6, 630 (2022)).

SKILLS

  • Deep learning frameworks and runtimes: PyTorch, JAX, vLLM, TensorRT, MIGraphX, ROCm, ONNXRuntime, QNN, TIDL
  • Model optimization: pruning, knowledge distillation, neural architecture search, quantization (PTQ, QAT, SpinQuant, AWQ, GPTQ)
  • Hardware: NVIDIA Orin & Thor, AMD Strix Halo, Qualcomm HTPs & DSPs, Samsung Exynos, TI TDA4, ARM CPUs
  • Frameworks & software: CI/CD, GitHub Actions, Sphinx, Docker, Kubernetes, SLURM
  • Scientific ML: Bayesian inference, Gaussian processes, optimization on manifolds, quantum simulation, data visualization

OPEN SOURCE

  • gd-qpt: Gradient descent quantum process tomography (with JAX), 2022
  • qst-cgan: Quantum state tomography with conditional GANs (with TensorFlow), 2022
  • qutip-tensorflow: Tensorflow backend for QuTiP (Google Summer of Code project mentor), 2021
  • qgrad: Quantum simulation in JAX for autodiff, (developer and mentor for Google Summer of Code project), 2020
  • matsubara: Hierarchical Eq. of motion solver for quantum dynamics, 2020 make-your-code-count: A guide to building open-source scientific computing projects (co-developer), 2019
  • pennylane: Differential programming of quantum circuits (contributed to core library, developed many tutorials and demos for applications), 2019
  • piqs: Permutational Invariance Quantum Solver for Lindblad open quantum systems (co-developer), 2018
  • dipy: Developed the IVIM module as a Google Summer of Code project, 2016

TALKS AND PRESENTATIONS

AWARDS & ACHIEVEMENTS

  • 2022: MITACS Accelerate fellow, Xanadu and UWaterloo, Canada
  • 2017: International Program Associate, RIKEN, Japan
  • 2016: Summer Research Fellow, Indian Academy of Sciences, Bengaluru
  • 2013: Selected for merit scholarship (NEST), Center for Excellence in Basic Sciences, Mumbai

TEACHING

  • Chalmers University of Technology — teaching assistant (2019–2022) for Master’s-level courses: Learning from data (TIF285), Applied machine learning, Advanced simulation and machine learning, Deep machine learning and reinforcement learning, C programming.
  • Master’s thesis supervisor (2020): quantum capsule neural networks, thesis by an Erasmus Mundus student (KU Leuven), co-supervised with Dr. Anton Frisk Kockum.
  • BITS Pilani, K.K. Birla Goa Campus — teaching assistant, Quantum Mechanics I (2016).

PUBLICATIONS

See Google scholar for updated list of publications.

  1. Tranheden W, Ahmed S, Dubhashi D, Matthiesen J, von Essen H. “FlashHead: Efficient Drop-In Replacement for the Classification Head in Language Model Inference.” arXiv:2603.14591 (2026).

  2. Lambert N, Giguère E, Menczel P, Li B, Hopf P, Suárez G, Gali M, …, Ahmed S, et al. “QuTiP 5: The quantum toolbox in Python.” Physics Reports 1153, 1-62 (2026). arXiv:2412.04705.

  3. Gaikwad A., Torres ME, Ahmed S, Kockum AF. “Gradient-descent methods for fast quantum state tomography.” arXiv:2503.04526 (2025).

  4. Recio-Armengol E., Ahmed S, Bowles J. “Train on classical, deploy on quantum: scaling generative quantum machine learning to a thousand qubits.” arXiv:2503.02934 (2025).

  5. Bowles J, Ahmed S, Schuld M. “Better than classical? The subtle art of benchmarking quantum machine learning models.” arXiv:2403.07059v2 (2024).

  6. Ahmed S, Killoran N, Carrasquilla J.F. “Implicit differentiation of variational quantum algorithms.” arXiv:2211.13765 (2022).

  7. Ahmed S, Quijandría, F., & Kockum, AF. “Gradient-descent quantum process tomography by learning Kraus operators.” Phys. Rev. Lett. 130, 150402 (2023). arXiv:2208.00812.

  8. Ahmed S, Muñoz CS, Nori F, Kockum AF. “Quantum State Tomography with Conditional Generative Adversarial Networks.” Phys. Rev. Lett. 127, 140502 (2021). arXiv:2008.03240.

  9. Ahmed S, Muñoz CS, Nori F, Kockum AF. “Classification and reconstruction of optical quantum states with deep neural networks.” Phys. Rev. Research 3, 033278 (2021). arXiv:2012.02185.

  10. Li B, Ahmed S, Saraogi S, Lambert B, Nori F, Pitchford A, Shammah N. “Pulse-level noisy quantum circuits with QuTiP.” Quantum 6, 630 (2022). arXiv:2105.09902.

  11. Lambert N, Ahmed S, Cirio M, Nori F. “Modelling the ultra-strongly coupled spin-boson model with unphysical modes,” Nature Communications 10, 3721 (2019). arXiv:1903.05892.

  12. Shammah N, Ahmed S, Lambert N, Liberato S. D., and Nori F. “Open quantum systems with local and collective incoherent processes: Efficient numerical simulations using permutational invariance,” Physical Review A 98 (6), 063815 (2018).

  13. Bergholm V, Izaac J, Schuld M, Gogolin C, Alam MS, Ahmed S, et al. “PennyLane: Automatic differentiation of hybrid quantum-classical computations,” arXiv:1811.04968.