Habeel Ali

Embedded Systems & HPC Engineer | CUDA, Rust, RTOS

Based in: Lahore, PakistanCurrently: Embedded Systems Engineer, Powersoft19Experience: 1 year

Hire me

Computer Engineer specializing in high-performance computing and embedded systems. Proven track record in optimizing CUDA MoE engines (5.8x speedup) and developing no-allocator RTOS with guaranteed nested interrupt latency (5.4µs). Experienced in hardware-software co-design, RISC-V kernel autotuning, and industrial firmware optimization.

Experience

Embedded Systems Engineer

Powersoft19LahoreworkJul 2026 – Present

Summary

Develop and debug Zynq-7000 firmware for industrial power-electronics equipment; trace faults in an existing codebase, improve timing-sensitive paths, and verify fixes on physical hardware.

Results

  • Reduced main control interrupt execution time by 33% (from 36us to 24us) under load conditions.
  • Optimized lwIP driver ISR, achieving a 75% reduction in execution time (from 8us to 2us).
  • 33% reduction in main control interrupt latency
  • 75% reduction in lwIP ISR execution time
Zynq-7000FirmwareHardware Verification

DevOps and Full-Stack Engineer

IT-vate SolutionsLahoreworkJun 2025 – Jun 2026

Summary

Maintained a React, Vite and Express business application; set up GitHub build workflows and Vercel preview deployments for review before merging. Managed production hosting, domain configuration, SSL and Cloudflare; integrated Firebase and SMTP-based email functionality.

Results

  • Configured Cloudflare DDoS protection and managed production hosting for a business application targeting the Pakistani market.
  • Maintained infrastructure supporting 100 concurrent users with automated CI/CD via GitHub Actions and Vercel.
  • Supported 100 concurrent users
  • Implemented DDoS protection and SSL security
ReactViteExpressGitHub ActionsVercelCloudflareFirebaseSMTPSSL

Skills

Technical

CUDA
PyTorchPyTorch
C++C++
RustRust
PythonPython
Automated testing
CC
ReactReact
API integration
scikit-learn
cuBLAS
CUDA Graphs
INT8 quantization
Model evaluation
LinuxLinux
GitHub ActionsGitHub Actions
QEMU
Profiling
CloudflareCloudflare
DockerDocker
ViteVite
ExpressExpress
Node.jsNode.js
FirebaseFirebase
JavaScriptJavaScript
VercelVercel
Transformers
Verilog
SMTP
VHDL

Projects

CUDA Mixture-of-Experts Inference Engine

Lead Developerpersonal
Implemented routing, token gathering, SiLU gating, scattering and weighted aggregation kernels for Phi-tiny-MoE-instruct, using PyTorch/cuBLAS for expert matrix multiplication and conditional CUDA graphs for dispatch. Reduced measured 256-token, batch-one prefill latency from 221.76 to 38.03 ms (5.82x) against the original model path; kept estimated GPU use below a 24 GiB ceiling. Collected 1,890 model timings and profiler traces; passed 16,873 numerical and routing checks with zero measured error in the tested setup.
C++CUDAPyTorchPythoncuBLASCUDA Graphs

RISC-V INT8 Kernel Generator and Autotuner

Lead Developerpersonal
Wrote a CLI that generates convolution, depthwise, pointwise and fully connected INT8 kernels, cross-compiles candidates, checks results in QEMU and searches under flash and SRAM limits. Exported standalone no_std kernels and C headers. On ESP32-C3, measured 1.45x and 1.75x faster complete-model inference than the fastest tested C baselines for two pretrained models. Verified 480 hardware model runs and 2,256 intermediate tensors across three hard-reset sessions with no numerical mismatches.
RustRISC-VQEMUCESP32-C3

Battery State and Health Estimation

Collaboratoracademic
Trained Random Forest models on NASA battery datasets using cell-out validation and training-only scaling; deployed models under 75 KB to ESP32-S3. Collaborative project.
Pythonscikit-learnC++ESP32-S3

Rivet Real-Time Operating System

Lead Developerpersonal
Developed a no-allocator RTOS with preemptive and async execution, priority inheritance and SMP; tested on seven boards across three architectures with host tests, QEMU regressions and CI.
RustARMRISC-VXtensaQEMU

Education

BS Computer Engineering

COMSATS University Islamabad, Lahore CampusComputer EngineeringJul 2026
Hire HabeelReach out about a role, a contract or a conversation.For recruiters

Habeel's twin is AI, it can make mistakes.

More people on Kanvis

Browse more peopleCreate your own Kanvis