PORTFOLIO / GPU & C++ / MACHINE LEARNING / LANGUAGES / PROJECTS / LINKEDIN / CONTACT
A live software demonstration from my professional portfolio.
I'm a Geospatial Software Engineer and Cesium Certified Developer based in Abuja, Nigeria, working across satellite intelligence, enterprise GIS, GeoAI, and full-stack applications. At Proforce Galaxies Limited, I build geospatial software, develop GeoAI and GenAI pipelines, and automate spatial workflows for the defence, space, and satellite sectors.
My work runs from a satellite band or a field survey all the way to a database, an API, and an interface someone can actually use. That has meant mapping electricity networks, modelling solar-site suitability, analysing urban heat, and building 3D tools in the browser.
5+ years in geospatial work. B.Tech. in Remote Sensing & GIS, FUTA. Founder of GeoIntel.
“GIS and Remote Sensing can solve any problem, as long as it exists on Earth.”
— The line I keep coming back to.
Dedicated GPU compute, shader, and C++20 host-runtime repositories covering CUDA, WebGPU/WGSL, GLSL 4.60, and Modern C++20—each backed by automated GitHub Actions CI (nvcc, wgsl_reflect, glslangValidator / SPIR-V, GCC/Clang -Werror, and ASan/UBSan).
| Repository | Stack | What is implemented | CI & code |
|---|---|---|---|
| Flash Kernel CUDA | CUDA 12.x C++17 CMake |
Progressive SGEMM ladder (naive → +1 bank-conflict-padded 32×32 SMEM tile → 2D register-blocked 64×64×8 float4), fused RMSNorm + SwiGLU, single-pass online safe softmax, FlashAttention-style SRAM-tiled SDPA with causal masking, Blelloch scan, Nsight Compute profiling script, and analytical roofline modeling. |
Kernels · Tests |
| WGSL Compute Lab | WebGPU WGSL TypeScript |
Pure WebGPU compute workbench: 16×16 workgroup-tiled SGEMM, conflict-free multi-pass Blelloch prefix sum, two-stage parallel tree reduction, 65K+ ping-pong particle simulation + instanced renderer, GPUQuerySet timestamp profiling, and wgsl_reflect AST/binding contract tests. |
WGSL Shaders · Tests |
| GLSL Compute Engine | GLSL 4.60 OpenGL 4.6 C++20 |
9 GLSL 4.60 compute & graphics shaders: 3D Simplex curl-noise particles, GPU-driven AABB frustum culling (DrawElementsIndirectCommand + atomicAdd), parallel bitonic sort, 16×16 Forward+ tiled light culling, Cook-Torrance PBR deferred lighting, SDF raymarcher, and C++20 host runtime with compile-time std140/std430 ABI checks. |
Shaders · C++ Host |
| GPU Runtime C++ | C++20 Lock-Free AVX2 |
Zero-dependency host-side GPU runtime: 32-bin coalescing aligned memory pool, cache-line-isolated (alignas(64)) lock-free SPSC command ring buffer (acquire/release), multi-stream DAG task-graph scheduler with Kahn cycle detection & Graphviz DOT export, strided TensorView<T> + BLIS-blocked SGEMM, and chrome://tracing JSON profiler. |
Headers · Tests |
CUDA 12.x (SM 80/86/89) · WebGPU / WGSL · GLSL 4.60 / SPIR-V · Modern C++20 · CMake · GitHub Actions CI.
My ML work connects Earth observation with the engineering needed to trust an experiment: knowing which images were used, choosing a checkpoint without fitting to test labels, checking numerical correctness, and measuring what an inference change actually costs.
| Project | What is implemented | Evidence to inspect |
|---|---|---|
| GeoVision Bench | A from-scratch PyTorch land-cover classifier, image-hash split provenance, a colour-statistics baseline, validation-only calibration, and a paired augmentation study. | Experiment reports · Model card |
| TileFlow Inference | Halo-aware tiled inference with whole-image parity checks. A small CNN trained on EuroSAT images adds a real-image denoising workflow and a measured CPU memory/latency comparison. | Trained-model experiment · Numerical tests |
| ML Eval Forge | NumPy model training, calibration and a replay validator that rejects inconsistent saved models, predictions, metrics and loss history. | Debugging case study · Failure-path tests |
Python · PyTorch · NumPy · pytest · reproducible CLI workflows · GitHub Actions. These are focused portfolio experiments developed with coding-agent assistance. Each repository separates measured results from limitations; the EuroSAT image split does not establish performance in a new geographic region. The case study shows the debugging process, including a failing regression before the repair.
Python is where much of my geospatial and GeoAI work happens: raster processing, spatial analysis, data pipelines, APIs, and automation. TypeScript and JavaScript bring that work into interactive maps and applications. SQL with PostgreSQL/PostGIS handles the spatial data underneath.
I also write Rust, Go, C++, and Java. My public systems projects cover search, messaging, storage, gateways, and workflow execution:
| Language | Where you can see it |
|---|---|
| Python | GeoVision Bench: reproducible PyTorch training; TileFlow: numerical inference systems; UHI-LST: Landsat processing and analysis |
| TypeScript | Drone Flight Path Visualizer: React, CesiumJS, and CZML; portfolio source |
| JavaScript | Notification service and browser application work |
| SQL | PostgreSQL/PostGIS spatial data workflows, including the customer enumeration work in my portfolio |
| Rust | Strata Queue: partitioned messaging; Vortex Search: inverted indexes and BM25 ranking |
| Go | Zenith Gateway: rate limiting and circuit breakers; Sentinel KV: sharding, TTL, and write-ahead logging |
| CUDA / WGSL / GLSL | Flash Kernel CUDA: SGEMM, FlashAttention-style SDPA, RMSNorm/SwiGLU, scan; WGSL Compute Lab: WebGPU compute workbench; GLSL Compute Engine: GPU-driven OpenGL 4.6 compute & PBR shaders |
| C++17 / C++20 | GPU Runtime C++: aligned memory pool, lock-free SPSC queue, DAG scheduler; Hydra LSM: MemTables, SSTables, Bloom filters, and WAL |
| Java | Nexus Workflow: DAG scheduling, asynchronous execution, and retry policies |
3rd Place · Esri Young Scholars Award, Nigeria · 2025
My solar-site suitability work evaluates locations in Akure and Ikorodu using GIS-based multi-criteria analysis, weighted overlays, and conditional tools. It connects raster analysis to a practical energy-planning question: where should solar installations go?
ArcGIS Pro · MCDA · Raster analysis
| Work | The problem and the implementation |
|---|---|
| Electricity network assets | Published 11kV and 33kV network assets across five business hubs in Osun State through ArcGIS Enterprise Portal. |
| Customer enumeration | Field data and CIN collection for billing and network planning, using Field Maps, Survey123, and PostGIS. |
| Interactive 3D City Mesh | Integrated mesh scenes with weather, flood toggles, and lighting, using Three.js, CesiumJS, and ArcGIS. |
| TaxNaija | A Nigerian tax-compliance application with calculation engines and an AI assistant, built with React, TypeScript, Gemini, and Vercel serverless functions. |
| Land cover & environmental monitoring | Supervised LULC classification in Akure and satellite-based oil-spill assessment in Oyigbo, Rivers State. |
| Distributed data analysis | Sales reconciliation across eight regions, using Python, pandas, and a multi-agent / MapReduce architecture. |
View the full project collection ↗
| Discipline | Tools I use |
|---|---|
| GPU compute & shaders | CUDA 12.x · WebGPU · WGSL · GLSL 4.60 · OpenGL 4.6 DSA · SPIR-V · C++20 · CMake · Nsight Compute · glslangValidator · ASan/UBSan |
| Earth observation & GIS | ArcGIS Pro · ArcGIS Enterprise · QGIS · Google Earth Engine · ESA SNAP · ENVI · ERDAS · GDAL · GeoServer |
| GeoAI & scientific computing | PyTorch · NumPy · scikit-learn · ArcPy · GeoPandas · Rasterio · Random Forest · SVM · Hugging Face |
| ML evaluation & tooling | Seeded experiments · validation-only calibration · numerical parity · CPU/RSS benchmarks · pytest · coding agents · GitHub Actions |
| Web maps & 3D | CesiumJS · CZML · Three.js · Leaflet · Mapbox GL JS · OpenLayers · deck.gl · ArcGIS Maps SDK |
| Applications & APIs | React · Next.js · Django / GeoDjango · FastAPI · Node.js · Vite · Tailwind CSS · REST |
| Data & delivery | PostgreSQL · PostGIS · spatial ETL · Docker · Vercel |
| Enterprise & field GIS | Field Maps · Survey123 · Experience Builder · Dashboards · CityEngine |
| Generative AI | LLM integrations · Gemini API · GenAI-assisted geospatial workflows |
| When | Role | Organisation |
|---|---|---|
| 2025–present | Geospatial Engineer & Developer | Proforce Galaxies Limited, Abuja |
| 2025–present | Founder & CEO | GeoIntel |
| 2024–Jun 2025 | GIS Analyst | Ibadan Electricity Distribution Company |
| 2023–present | Freelance GIS Consultant | Upwork / Fiverr |
| 2022 | GIS Intern | Ibadan Electricity Distribution Company |
| 2020–2022 | Geospatial Analyst | Geotechnics Services Limited |
Education: B.Tech., Remote Sensing & GIS, Federal University of Technology, Akure · 2018–2024 · Magna cum laude.
More about my experience / Curriculum vitae
REA — contributed a fix for trailing-slash package exports, including regression coverage. I also published Windows browser reproduction results to help investigate an unresolved report.
My systems repositories above are another part of the picture: a place to work on the mechanics underneath applications—storage, concurrency, messaging, search, and execution.
- Esri Young Scholars Award, Nigeria — 3rd Place · 2025, for solar-site suitability modelling.
- Cesium Certified Developer · 2026.
- Certified Geospatial Engineer/Developer, micro1 · 2026.
- NASA ARSET: Hyperspectral Data for Land and Coastal Systems · 2026.
More training & continuing education
- Esri, 2025: Going Places with Spatial Analysis; GIS for Climate Action; Make an Impact with Modern Geo Apps; Power Data-Driven Decisions with ArcGIS Dashboards; Designing GIS Applications with Accessibility.
- Quantic, 2025: Decision-Making with Analytics; Business Intelligence in Action.
- Alison, 2022–2024: Advanced Diploma in GIS; Representing the Real World in GIS; GIS Data Quality and GPS.
I teach GIS and remote sensing, share work through talks and workshops, and write about how geospatial systems are changing.
- Watch: Geospatial FM — Power Utilities & Earth Observation Innovation
- Read: GenAI & LLMs in Geospatial Workflows
- Learn with me: GIS and remote sensing tutorials
- Explore: Talks, conferences, and other media
Table tennis. Football. Chess. I follow global affairs and geospatial research, and I like having something new to get better at.
If you're working on Earth observation, GeoAI, a spatial product, or enterprise GIS, let's talk.
Original material in this repository is available under the Noncommercial Use and Commercial Permission License. Noncommercial use is free. Commercial use requires prior written permission and a separate written agreement specifying the scope and agreed fee. Contact work.abdulfatai@gmail.com with the repository name and intended use.
Portfolio review and recruitment evaluation are permitted without a fee. Third-party components retain their own licenses. Permissions already granted for earlier versions remain valid; the new terms do not revoke those grants. These are custom source-available terms, not an OSI-approved open-source license.



