Brent A. Thorne

Brent A. Thorne

Embedded Systems Engineer • Scientific Computing Specialist • R&D Platform Architect
Firmware + Drivers • Real-Time Sensing • HPC Optimization • Applied AI

I design and build the low-level software that makes hardware intelligent — from kernel drivers and real-time sensor pipelines to GPU-accelerated scientific computing. I’m most at home where firmware meets physics and where careful engineering turns complex systems into reliable platforms.

What I Do

I build intelligent, resilient systems at the intersection of embedded firmware, real-time sensing, and scientific computing. My work focuses on:

I specialize in projects requiring low-level insight and high-level systems thinking.

Featured Projects

Below are project cards with architecture diagrams (Mermaid) to illustrate system design and data flow.
Repositories are private but available upon request.

Spectral Radar Decomposition Driver (Kernel-Space)

Developed a Linux kernel driver for a radar module streaming high-rate ADC data via DMA into real-time FFT engines.

Impact:
Enabled low-latency RF signal capture and spectral decomposition for imaging and diagnostics.

Results:
Zero-copy DMA transfer via mmap ring buffer, sustaining continuous ADC streaming without frame drops at full sample rate.

Tech:
C, Kernel Modules, DMA, mmap, FFTW

Repo: Private — available upon request

flowchart LR
    ADC[High-Rate ADC] --> DMA[DMA Engine]
    DMA --> KMOD[Kernel Driver]
    KMOD --> RB[Ring Buffer mmap]
    RB --> FFT[User-Space FFT Engine]
    FFT --> APP[Analysis Application]

Sphere Point Distribution for Agent Initialization

Generated uniform and quasi-uniform spherical distributions using Fibonacci, geodesic, and Lissajous curves with CUDA-accelerated force modeling.

Impact:
Optimized initialization for agent-based simulations and volumetric analysis.

Results:
Achieved near-uniform point distributions on the sphere with GPU-accelerated electrostatic relaxation, orders of magnitude faster than CPU-only baselines.

Tech:
CUDA, Python, AVX2, Matplotlib

Repo: Private — available upon request

flowchart TD
    INIT["Initialize Points on Sphere"] --> FORCE["Compute Repulsive Forces (CUDA)"]
    FORCE --> UPDATE["Update Positions"]
    UPDATE --> CHECK["Convergence Check"]
    CHECK -->|Not Converged| FORCE
    CHECK -->|Converged| OUTPUT["Final Distribution"]

Multi-Camera Real-Time Pipeline (V4L2 + RTSP)

Built synchronized multi-camera acquisition with low-latency streaming via MediaMTX.

Impact:
Supports embedded vision applications on constrained compute platforms.

Results:
Synchronized multi-stream capture with frame-accurate alignment, delivered at low latency over RTSP for downstream processing.

Tech:
V4L2, FFmpeg, MediaMTX, GStreamer

Repo: Private — available upon request

flowchart LR
    CAM1[Camera 1] --> V4L2[V4L2 Capture]
    CAM2[Camera 2] --> V4L2
    V4L2 --> SYNC[Frame Sync Module]
    SYNC --> ENCODE[FFmpeg Encoder]
    ENCODE --> RTSP[MediaMTX RTSP Server]
    RTSP --> CLIENT[Client Viewer / Processing Node]

Disaster Image Severity Classification (CNNs)

Fine-tuned convolutional models to classify damage severity from crisis imagery.

Impact:
Explored multimodal cues for humanitarian response modeling.

Results:
Compared fine-tuned CNN architectures (ResNet, EfficientNet) with transfer learning to classify damage severity across multiple disaster categories.

Tech:
PyTorch, TensorFlow, NumPy, OpenCV

Repo: github.com/fractalclockwork/Data200

flowchart LR
    IMG[Input Image] --> PREP[Preprocessing]
    PREP --> CNN[Convolutional Backbone]
    CNN --> FC[Dense Layers]
    FC --> OUT[Severity Classification]

Technical Focus Areas

Education

Experience

Freelance Research Engineer (2018–Present)

Senior HW/SW R&D Engineer — Grid Net (2013–2018)

Design Engineer — Embedded Systems Consultant (2011–2013)

Systems & Integration Engineer — OpenTV (2004–2011)

Patents

Contact

“Simple systems can grow complex behavior. Complex systems require careful simplicity.”