Skip to content
View alebal123bal's full-sized avatar

Block or report alebal123bal

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
alebal123bal/README.md

Alessandro Balzan

FPGA Engineer | Hardware Acceleration | Edge AI

I bridge the gap between PyTorch and Verilog/VHDL.

I live in the intersection of cloud and logic gates. I specialize in taking heavy computational problems (Computer Vision, AI, Optimization) and crushing them into constrained hardware with deterministic latency.

  • Focus: High-throughput data paths, RTL design, and hardware-aware AI.
  • Mindset: If it takes more than a microsecond, it's too slow.

⚡ Featured Projects

The Viral One. A zero-copy, hardware-accelerated pipeline on the Luckfox Pico.

  • Performance: Achieved 30 FPS on a tiny, $25 embedded Linux board.
  • Tech: C++, RGA Hardware Acceleration, MAVLink.
  • Impact: 19x speedup over OpenCV. Went viral on LinkedIn.

The CEO's Favorite. Real-time video processing on the Gowin Tang Primer 20K.

  • Tech: Verilog, HDMI, USB 2.0 High Speed.
  • Impact: Publicly praised by the CEO of Gowin Semiconductor for technical execution.

The Math. Custom implementations of ML algorithms from scratch.

  • Tech: Python, Numba (JIT compilation).
  • Why: I optimized the kernels for raw speed and lightweight usage.

🛠 The Stack

Domain Tools & Technologies
Hardware (RTL) VHDL, Verilog, SystemVerilog, Vivado, Gowin IDE
Software & AI Python (PyTorch, Numba), C/C++, Bash, Linux
Interfaces AXI4, Avalon Stream, UART, I2C, SPI
Optimization Bayesian Optimization (BOTorch), Multi-objective Search

💼 Experience Highlights

Leonardo S.p.A (via Overtech Consulting srl)

  • Role: FPGA Engineer (Consultant)
  • Focus: High-reliability RTL design for defense/aerospace applications.
  • Work: High bandwidth video processing.
    • Note: Specific project details are confidential.

Infineon Technologies

  • Role: R&D Intern
  • Work: Developed automated circuit sizing algorithms using Bayesian Optimization. Reduced design evaluation time significantly using symbolic analysis and math-heavy optimization.

♟️ Out of Office

  • Chess: FIDE Rating 2000. I play competitive tournaments. I like strategy.
  • K1 Kickboxing: Disciplined training. I like impact.


Disclaimer: All projects and code on this repository are personal, open-source endeavors developed on personal time using commercially available hobbyist hardware. They are not affiliated with, endorsed by, or related to the proprietary technologies of my current employer or clients.

Pinned Loading

  1. khadas_yolov8n_multithread khadas_yolov8n_multithread Public

    Real-time YOLOv8n UAV detection at the sensor's 46 FPS ceiling, in ~140 MB of RAM

    C++ 32 1

  2. luckfox_UAV_detection luckfox_UAV_detection Public

    Real-time drone detection @ blazing fast 30 FPS on RV1106 | Zero-copy DMA + hardware RGA acceleration | YOLOv5 RKNN | MAVLink telemetry streaming | Picoclaw integration

    C++ 29 6

  3. Gowin_FPGA_CV Gowin_FPGA_CV Public

    FPGA-based camera streaming on Gowin Tang Primer 20K | OV5640 @ 640×480 @ 51.45 FPS | HDMI + USB 2.0 HS

    Verilog 7

  4. RKNN_TRAIN_YOLO RKNN_TRAIN_YOLO Public

    Entire pipeline for training, converting and exporting a YOLO model into RockChip NPU

    Python 1

  5. RKLLM_LLAMA_QWEN RKLLM_LLAMA_QWEN Public

    Entire pipeline for running optimized LLM models on the RK3588S, both on the NPU (RKLLM) or the CPU (llama)

    Shell

  6. Latex_TurboJet_boilerplate Latex_TurboJet_boilerplate Public

    Latex boilerplate and detailed instructions for fast PDF production.

    TeX