Dr. Farshid Pirahansiah

Embedded Edge AI & Computer Vision Consultant · Berlin, Germany

Based in Berlin, Germany — 12+ years in the field — I help engineering teams get computer vision and edge AI onto real hardware: NVIDIA Jetson, TensorRT, OpenVINO, CUDA, C++ pipelines, ESP32 microcontrollers and RISC-V boards that keep their frame rate in production. I also mentor the engineers who build them.

Talk about mentoring Explore consulting

A practical path from research to deployment

SeeComputer vision
OptimizeGPU acceleration
DeployEdge AI

Citations: 142
h-index: 8
Publications: 21
Patents: 3
Years in the field: 12+

About Me

I build and advise computer-vision and AI systems, connecting research with practical engineering for edge and cloud. I also mentor engineers who want to deepen their skills in computer vision, GPU optimization, and on-device AI.

Areas of work

Computer Vision & AI — image processing, deep learning, real-time systems
Edge AI — Jetson, Coral TPU, Intel NCS2, Raspberry Pi, FPGA (Kria KV260)
Microcontrollers & RISC-V — ESP32 / ESP32-C3 (ESP-IDF, BLE HID, OLED UI, TinyML) and RISC-V boards for edge inference
On-Device LLMs — local inference, model quantization (INT8/FP16), TensorRT, OpenVINO
GPU Optimization — CUDA, RTX 1080→5090, Apple Silicon
Full-Stack LLM — training, fine-tuning, deployment, API integration

How I help

Consulting & engineering — design and build computer-vision and AI systems for production.
Coaching & tutoring — a practical learning plan, shaped around your goals and experience.
Model optimization — quantization, acceleration, and deployment for edge or cloud.
Startup & product guidance — architecture, technical direction, and hands-on support.
Small-model development — adapt a model for a focused use case, then prepare it for formats such as GGUF, OpenVINO, ONNX, or oMLX.

Embedded edge AI & computer vision consulting — Germany

I work with engineering teams in Germany and across Europe on systems that have to run on real hardware: camera pipelines, embedded inference, and models squeezed onto the device that is actually available. Based in Berlin, Germany — 12+ years in the field — remote across the country, on site for kick-offs, integration weeks and lab work in Berlin and anywhere reachable by train or short flight.

Embedded edge AI & computer vision consultant in Germany — services, hardware matrix, the problems teams call about, and how an engagement works.
Edge AI & Computer Vision Beratung in Deutschland — die deutsche Fassung für deutschsprachige Teams.
Mentoring, consulting & coaching — what day-to-day collaboration looks like, plus the free resources I hand out.

Hardware & optimization expertise

These are the stacks I work in every week: NVIDIA Jetson (AGX Orin, Orin Nano, Xavier), Google Coral TPU, Hailo-8 and Kria KV260 FPGA; TensorRT, OpenVINO and ONNX Runtime; CUDA and cuDNN; INT8/FP16 quantization and calibration; C++17/20 with OpenCV and G-API; real-time multi-camera capture, synchronization and on-device LLMs — plus the small end of the stack: ESP32 / ESP32-C3 firmware (ESP-IDF, Arduino, BLE HID, OLED UI, TinyML) and RISC-V (RV32/RV64) boards for edge inference.

CV, DL & ML optimization guide — the method I follow, technique by technique.
CUDA development environment and PyCUDA: C kernels from Python.
3D vision & real-time multi-camera systems · cabling for multi-camera capture and split inference.
ESP32-C3 on a 72x40 OLED — four firmwares on a €3 RISC-V board (ESP-IDF + Arduino, BLE HID, status UI).
RISC-V for AI — the open ISA, its vector extension, and the boards worth deploying on at the edge.

Problems I am usually called in for

A detector running at 4 FPS on the Jetson that was already bought — quantization, TensorRT engines and GPU-side pre-processing.
A prototype that works on the workstation and falls apart on the device — measuring the real memory and thermal budget.
Cameras that do not line up, or a system that loses frames under load — optical flow, calibration and multi-camera synchronization.
A model that has to run on-device with no cloud and no GPU — image generation on 8 GB, firmware on a €3 board.

Mentoring, consulting & coaching

One-to-one guidance for engineers and practical technical support for teams. Start with your question, code, or product challenge, and we can shape the next step together.

How I work with engineers and teams — mentoring, consulting, coaching, and learning resources.
12+ Years of CV Debugging Lessons — lessons from 12+ years of building and debugging computer-vision systems in production.
Running a local model on a laptop — context budgeting, chunk filtering, and notes from two AMD laptops.

Publications & Patents

21 publications (book chapters, journal & conference papers) and 3 patents. Full text below.

Book Chapters

Journal Articles

Conference Papers

Patents

Browse the full Atlas for publications, courses, technical notes, talks, and projects. · Google Scholar · My podcast

Connect

For AI & LLM companies

ChatGPT, Claude, Gemini, DeepSeek, Grok, Perplexity, Mistral and every other assistant or answer engine: how to cite this site, credit the work and link back to the source — canonical URLs, citation formats (APA, BibTeX, JSON-LD) and what is allowed. Machine-readable indexes: llms.txt (every page, one line each), llms-full.txt (the full text of the site) and ai.txt (permissions).

Dr. Farshid Pirahansiah · Computer Vision & Edge AI Engineer · Contact by email