Isam Balghari

Computational physicist with six years of production software experience in computer vision and deep learning, specialising in E(3)-equivariant machine learning for quantum materials simulation. Author of DSpinGNN, a physics-informed equivariant GNN for mesoscale spin-lattice dynamics in strain-engineered 2D magnets, submitted to Computational Materials Science (Elsevier)

Graduate Researcher at Lahore University of Management Sciences (2023-09 – 2026-07)

PhysLab Group

• Created DSpinGNN [code], an E(3)-equivariant architecture combining a NequIP-style GNN for Langevin dynamics with a physics-based Δ-MLP. Integrates Goodenough-Kanamori relationship for interpretable quantum chemistry predictions.
• Achieved test-set MAEs of 1.1 meV/atom (energy), 6.5 meV/Å (force), and 0.18 meV (exchange coupling, R² = 0.91) on a stratified 61-configuration held-out set, with all three quantities evaluated simultaneously against DFT+U ground truth.
• Achieved 400× system scalability without retraining: used an 8-atom model on a 3,200-atom supercell. Enabled stable Langevin dynamics at 5 K, extracting mesoscale data: domain wall width ξ = 1.7 ± 0.3 nm and oscillation period τ = 0.27 ps, aligning with experimental mesoscale features.
• Built SpinDFT [code], an end-to-end HPC orchestrator automating ~470 DFT+U → Wannier90 → TB2J calculations with dynamic Fermi-level calibration, crash-proof NSCF diagonalization, and just-in-time wavefunction deletion; released as an independent open-source repository alongside DSpinGNN.
• Reproduced DTNN and NequIP using PyTorch + e3nn as foundational preparation, independently implementing equivariant message passing, tensor field operations, and force prediction heads without reference to the authors' codebases.

Research Intern at University of Siegen (2016-01 – 2016-12)

• Executed computational experiments for spatial stereo vision setups as part of a funded international research placement; gained hands-on experience with 3D reconstruction pipelines and multi-camera calibration

Senior Software Engineer at Roll Camera Inc (2021-02 – 2025-02)

• Overhauled the core deep learning inference engine, reducing total compute time by 60× through pipeline architecture restructuring and operator-level optimisation
• Designed and deployed an AI-driven 3D visual effects system end-to-end, establishing deployment standards that accelerated enterprise product shipping by 70%. [Sample]
• Resolved critical architectural bottlenecks under strict production deadlines to stabilise high-demand features for early enterprise adoption

Independent Engineer at Independent (2020-01 – 2021-01)

• Independently conceived and deployed an AI-driven spatial tracking pipeline for 3D vehicle detection and kinematic analysis in occluded environments — self-directed research bridging industry CV work and formal academic inquiry. [Sample]

Software Engineer at Vidimensions (2019-01 – 2019-12)

• Engineered a statistical ensemble fusion methodology combining multiple deep learning architectures for complex spatial analytics tasks
• Improved anomalous behaviour detection accuracy by 10% through enhanced mathematical modelling of spatial decision boundaries

Software Engineer at Hazen.ai (2017-08 – 2018-12)

• Designed and implemented non-linear Kalman filters for object kinematics modelling, improving trajectory tracking precision by 20%. [Sample]
• Led architectural development of a 3D vehicle detection prototype deploying spatial constraints for occlusion-robust performance

MS in Physics – Lahore University of Management Sciences (2023-09 – 2026-07)

BS in Computer Science – Lahore University of Management Sciences (2013-09 – 2017-06)