Millions of neurons, drawn at the speed of thought.
Real-Time Neural Visualisation·GPU / XR Engineering·Connectomics
Building for extended reality.
I build GPU-accelerated systems for visualising and analysing large-scale neural data. The NeuroConstellation platform family renders up to 5.28M neurons across 5 platforms (desktop, mixed reality, VR, and WebGPU), and I analyse full-scale CA1 connectomes (human and mouse) directly on the GPU.
I defended my PhD in Life Sciences and Biotechnologies at the University of Sassari, Italy, on 6 May 2026 (degree conferred 16 June 2026). My doctoral research developed the NeuroConstellation platform family, five GPU-accelerated visualisation tools for rendering large-scale neural network activity across desktop, mixed reality, virtual reality, and web-browser environments. Alongside the platforms, I analysed full-scale single-cell CA1 connectomes of the human (~5.28M neurons) and mouse (~288K) on the GPU, producing one of the first matched cross-species comparisons at this scale (manuscript in preparation) and identifying a conserved GABAergic neurogliaform rich-club.
My work sits at the intersection of computational neuroscience, extended reality, and human-computer interaction. I collaborate with researchers at the Institute of Biophysics CNR Palermo (Migliore lab), EBRAINS-Italy, and Aix-Marseille University, where I completed a 6-month international research mobility from August 2025 to January 2026. My ongoing research includes EEG signal analysis using MATLAB and brain co-simulation frameworks within the EBRAINS-Italy infrastructure.
Before the PhD, I earned an M.S. in Software Engineering from UESTC, China, where my thesis focused on fine-grained emotion detection from text using deep learning, and a B.S. in Computer Science from the University of Sindh, Pakistan. This interdisciplinary background, combining software engineering, NLP, and neuroscience, shapes how I approach complex visualisation challenges.
Notable distinction: Part of the first group of international students invited to serve as museum guides in UESTC's 60-year history (2019), delivering special lectures on Chinese technological advancement to a global audience.
Outside research, I have been actively involved in community work, most notably as an ILM Ambassador with the British Council's Take a Child to School programme, which contributed to enrolling over 110,000 out-of-school children across Pakistan.
One GPU pipeline carries full-scale brain data from raw connectome to a frame you can explore, on any screen.
Five GPU-accelerated visualisation systems spanning desktop, mixed reality, virtual reality, and the open web, rendering neural activity at scales previously inaccessible to interactive exploration.
Draw a digit and a real neural network classifies it as you draw, shown as the small network from 3Blue1Brown's series: a 784-16-16-10 perceptron with every neuron and weighted connection on screen. The inference runs in raw WebGPU compute shaders, and the activations lighting the neurons are read back from the same GPU buffers that draw them. Trained on real MNIST (95.8% test accuracy, about 13,000 parameters), with no machine-learning or graphics libraries. Educational, not state of the art: deliberately the small, legible network where every weight is visible.
Live demo SourceGPU-accelerated desktop visualisation of large-scale neural network activity, including human hippocampal CA1 datasets. Six visualisation modes alongside parallel GPU-accelerated analysis (firing-rate maps, recruitment maps, spike-history trails); validated against a Python NumPy/SciPy reference pipeline.
Mixed reality visualisation of mouse hippocampus CA1 neural activity on Microsoft HoloLens 2, using a distributed workstation–headset architecture for high-resolution rendering. Hand gesture interaction via Microsoft MRTK, with a novel proxy container interaction paradigm for manipulating dense neural datasets.
Stereoscopic VR visualisation of human hippocampus CA1 neural activity in Varjo XR-4 with Ultraleap hand tracking. Configurable dataset loading and confidence-based gesture interaction enabling six-degree-of-freedom manipulation in immersive space.
Browser-based neural visualisation using WebGPU compute shaders. All six visualisation modes translated from HLSL to WGSL. Zero installation, runs in Chrome, Edge, Firefox, and Safari on any WebGPU-capable browser.
Open live demoA hand-built WebGPU globe visualising live USGS earthquake data: ~9,000 events rendered as depth-coloured, magnitude-scaled spikes in a single instanced GPU draw call. Written from scratch against the raw WebGPU API (WGSL shaders, matrix math, arcball camera), with time-lapse playback, place search, and full accessibility. No three.js, no map SDK, no build step.
Live demo SourceA zero-install web app that runs a real sentence-transformer (all-MiniLM-L6-v2) fully in the browser on WebGPU. It embeds ~2,400 live-fetched emotion-labelled tweets into 384-D space, projects them to 3D with hand-written PCA, and renders the cloud as an instanced WGSL point system. Type a sentence and it is embedded live on the GPU, placed by the same basis, and linked to its true cosine nearest neighbours with a k-NN emotion prediction. No three.js, no backend, no build step.
Live demo SourceA zero-install web app that replays real 64-channel motor-imagery EEG (PhysioNet) and decodes imagined left/right hand movement live in the browser. A hand-written in-browser FFT drives mu (8-12 Hz) and beta (13-30 Hz) band power; a WGSL scalp topographic map renders on a real 3D head mesh; and event-related desynchronisation over the C3/C4 motor cortex calls which hand, self-scored against the true cue. The same EEG/BCI analysis I run in MATLAB and MNE, rebuilt to run for anyone with a browser.
Live demo SourceA zero-install web app that pushes about a million particles (up to 4.2M) through one real hour of NOAA GFS 10 m wind, entirely in the browser on WebGPU. A hand-written WGSL compute shader samples the wind field under every particle and advects it across a world map, leaving glowing motion trails coloured by wind speed. Because the field is one steady forecast snapshot, every trail is a true streamline. No three.js, no backend, no build step.
Live demo SourceA raw WebGPU flythrough of the real Milky Way. Every point is an actual star from ESA's Gaia mission: a random sample of 1,013,200 stars with full 6D measurements. Their orbits were integrated on the CINECA Galileo100 supercomputer (a 34-task SLURM array, gala / astropy), compressed to about 112 bytes per star, and reconstructed live on the GPU, so you can scrub the Sun's neighbourhood across 500 million years in either direction and click any star to draw its orbit. A compute pass evaluates every orbit, an HDR render pass draws additive star sprites, and GPU picking finds the star under the cursor. Hand-written WGSL, measured error reporting, no rendering libraries.
Live demo SourceA privacy-first wedding photo gallery built for a conservative Pakistani family. Each guest finds their own photos by scanning their face on their own phone, with the recognition running entirely on the device, so no selfie is ever uploaded. Photos of the family's women ship AES-256 encrypted in the browser and open only with a personal code, letting even conservative families share an online gallery while respecting purdah. Features a three.js 3D memory wall, WebGL2 GPU petal particles, one-tap WhatsApp sharing, reels and videos, and a privacy-safe Cloudflare analytics dashboard. Deployed live: 2,740 photos across two weddings.
Live demoA private, invite-only family tree (a "shajra") in daily use by my own Mirani family, built for non-technical relatives adding people from their phones over WhatsApp, in Sindhi and Urdu. Every addition or edit is a change request reviewed by a family moderator before it enters the tree, with a full audit log. Names transliterate to Sindhi automatically, kinship ("how is X related to Y") is computed from the actual family graph rather than guessed, and the family owns its data through one-click GEDCOM 7 export. Row-Level Security enforces privacy and roles in the database itself, and the data model is already multi-tenant-ready. The real tree is never public, so the demo below runs the same code over a fictional family, with zero database access and no account needed.
Try the demo SourceClient-GPU browser tool for single-cell CA1 connectomes, cell types, guided wiring tours, the structural rich-club, and live spiking-activity playback. Two species instances: mouse (~288K neurons) and human (~5.28M), computed entirely in-browser via WebGPU with no server.
Mouse explorer Human explorerGraph-theoretic analysis of full-scale single-cell CA1 connectomes: one of the first matched human/mouse cross-species comparisons at this scale, a conserved GABAergic neurogliaform rich-club, and a collinearity-aware variance decomposition of structure-to-function predictability.
Interactive viewers that place a representative published reconstruction of each CA1 cell type (NeuroMorpho.org) at its real soma position and render the whole population on the GPU, a "neuron forest" at brain scale.
Morphology lab Neuron forestSignal processing pipelines using MATLAB for EEG data analysis, implementing machine learning algorithms for pattern recognition and brain state classification.
Co-simulation tasks integrating multiple brain modelling platforms to bridge different scales of neural organisation, in collaboration with EBRAINS-Italy.
Fine-grained emotion classification using LSTM, CNN, and transformer architectures. Research conducted at UESTC NLP Lab.
Final-year project officially adopted as the Sindh University Larkana Campus web portal. First prize in the departmental software project exhibition.
Five browser-native tools, each rendering on your GPU through WebGPU, no install, no headset. Best in Chrome or Edge 113+ on a desktop or laptop.
Live demos are available on request. The sites are temporarily offline for maintenance.
5.28M-neuron activity visualiser, six modes, all on WebGPU compute shaders.
Launch →Single-cell wiring explorer — 288K neurons, cell types, rich-club, activity playback.
Launch →The full human connectome — 5.28M neurons, explored entirely in the browser.
Launch →A representative published reconstruction per cell type, from NeuroMorpho.org.
Launch →Every neuron drawn as its real morphology at its true soma position.
Launch →Datasets: full-scale mouse (~288K) and human (~5.28M) CA1 scaffold models (Gandolfi, Mapelli, Solinas & Migliore; EBRAINS / EU Human Brain Project), NEST-simulated on the CINECA Galileo100 cluster.
A first-author body of work spanning the full NeuroConstellation toolchain, one Q1 journal paper (Scientific Reports, IF 3.9), eight first-author manuscripts (one under review at iScience, two in revision, five ready for submission), and three published conference abstracts.
Scientific Reports · Nature Portfolio
The Evaluation and Comparative Analysis of Role Based Access Control and Attribute Based Access Control Model
15th International Computer Conference on Wavelet Active Media Technology and Information Processing (ICCWAMTIP), Chengdu, China, pp. 35–39
DOI: 10.1109/ICCWAMTIP.2018.8632578NeuroConstellation: Interactive Visualisation and Analysis of Million-Neuron Datasets Using Game Engine Technology
iScience, Unity 3D platform rendering up to 5.28M neurons at interactive frame rates
HoloNeV: Advancing Human-Computer Interaction in Neuroscience Through Mixed Reality Neural Visualisation
MDPI Electronics, resubmission in progress
Single-Draw-Call Immersive Visualisation of 5.28 Million Neurons with Hand-Tracked Interaction in Video Pass-Through Mixed Reality
Varjo XR-4 — adding connectivity / activation visualisation
Mixed Reality Visualisation of Simulated Mouse Hippocampus CA1 Activity with Hand-Gesture Interaction on Microsoft HoloLens 2
NeuroConstellation Web: Client-Side WebGPU Compute for Interactive Visualisation and Analysis of Million-Scale Neural Activity
Target: IEEE Access
Live demoA Conserved GABAergic Neurogliaform Rich-Club Organises the Mammalian CA1 Connectome
First matched cross-species (human / mouse) single-cell connectome comparison
A Web-Based WebGPU Explorer for Single-Cell Connectomes of the Mammalian CA1
Mouse HumanDegree Dominates: A Collinearity-Aware Variance Decomposition of Structure-to-Function Predictability in a Full-Scale CA1 Connectome
HoloNeV: Holographic Visualisation Tool for Neural Network Activity
J. Comput. Neurosci. 54(Suppl. 1), S227 · CNS*2025, Florence (Poster P201)
Programme entryNeural Circuit Visualizer: A Web-Based Platform for the Simulation and Visualisation of Full-Scale Point-Neuron Circuits
J. Comput. Neurosci. 54(Suppl. 1), S361 · CNS*2025, Florence (Poster P335)
Programme entryHolographic Visualisation of Neural Network Activity
EBRAINS-Italy, Naples, Italy, GPU-accelerated indirect instancing on HoloLens 2
Neural Network Visualisation & AI
Educational Programme Management
Examination Quality Control
Machine Learning Research
Banking Technology Department
University of Sassari (UNISS), Italy
Thesis: Real-Time Visualisation of Large-Scale Neural Networks: GPU-Accelerated Methods for Desktop, Extended Reality, and Browser Platforms. [thesis repository]
Degree conferred at convocation: 16 June 2026.
Supervisors: Prof. Sergio M. G. Solinas and Prof. Paolo Enrico (Department of Biomedical Sciences).
International mobility (completed): Institut de Neurosciences des Systèmes (INS), Aix-Marseille Université, France (Aug 2025 — Jan 2026).
University of Electronic Science and Technology of China (UESTC)
Thesis: Fine-Grained Emotion Detection from Text using Deep Learning. Specialisation in Machine Learning, NLP, and Software Architecture.
University of Sindh, Pakistan
Focus: Software Development, Database Systems, Web Technologies. First prize in Final Year Software Project Exhibition.
New Banat High School, BISE Hyderabad
The Best Academy, BSE Karachi
Third prize for outstanding academic research performance
Second prize for excellence in arts, sports, and voluntary work
Recognition for leadership and academic excellence
First prize in Final Year software project exhibition
Awarded laptop for exceptional academic performance
Outstanding performance during bachelor studies
International Cycling Competition
Muslim Students Spring Sports, UESTC
UESTC Summer Sports
Muslim Students Spring Sports
Cultural Activities at UESTC
Electronic Museum Award, UESTC
Winter Inter-Dormitory Olympics
Organising seminars and cultural events
University of Electronic Science and Technology of China (UESTC)
Part of the first group of international students invited to serve as museum guides in UESTC's 60-year history. Delivered special lecture introducing Chinese technological advancement to a global audience.
Featured in UESTC NewsWorkshop organised by F. de Pasquale & S. Della Penna
Environmental & Computational Neuroscience, Aix-Marseille University
BRAin DYnamics Academy
School of International Education
UET Lahore & FAST University with IEEE
Build your Skills, Knowledge, Abilities (SKAs)
Sports & Youth Affairs Department, Government of Sindh
Statistical analysis training
Education Ambassador for ILM-POSSIBLE: Take a Child to School (TACS) programme.
Impact: Contributed to enrolment of 110,000+ out-of-school children across Pakistan's four provinces.
British Council programmeEnvironmental and cultural organisation
International Students Union elections management
Community-building activities
China Soft Conference 2018
40th Anniversary of China's Reform Forum
International Freshers' Orientation
Promotional interviews for international education
International Students Union initiative
Full contact details available on request.
Doctoral supervisor · Associate Professor
Dept. of Biomedical Sciences, University of Sassari, Italy
Doctoral co-supervisor · Full Professor
Dept. of Biomedical Sciences, University of Sassari, Italy
International mobility host · Senior Researcher
Institut de Neurosciences des Systèmes (INS), Aix-Marseille Université, France
Co-author · Professor & Senior Researcher
Institute of Biophysics, National Research Council (CNR), Palermo, Italy
Open to postdoctoral positions, research collaborations, and consulting opportunities in computational neuroscience, XR, and GPU-accelerated visualisation.
The live demos are temporarily down for maintenance. I'd be glad to share a live link or give you a walkthrough.
Email Safeer