M.Sc. · Electrical Engineering
Telecommunications Systems
Malek Ashtar University of Technology · Tehran, Iran
Thesis: improving the accuracy of man-made object detection in aerial and satellite images using deep learning.
AI & Computer Vision Engineer
Machine learning engineer with seven years of experience specializing in computer vision — from classical image processing to modern deep learning — with a growing focus on generative AI, large language models and multi-modal systems.
01 — About
I am a machine learning engineer with seven years of experience specializing in computer vision. My expertise spans classical image processing and modern deep learning methods, and I work hard to keep up with a field that never stands still.
Alongside my professional work, I actively study generative AI, particularly large language models (LLMs) and vision-language models (VLMs), and I build and release open-source tools that have drawn a lot of attention from the community. I am most excited about combining LLMs and vision models into multimodal solutions.
I put new AI tools to work quickly: when an open-source model is released, whether an LLM, ASR or TTS, it often becomes a usable offline product within weeks. DikteYar, ParSiGo and DocQAI all came out of that approach, and I rely heavily on AI agents such as Claude Code to build them.
I do my best work in dynamic, innovative environments, where I can collaborate with like-minded people, try out new ideas, and build things in machine learning and AI that didn't exist before.
02 — Experience
May 2024 — Present
Independent · Remote
Independent machine learning engineer and consultant — designing, training and shipping computer-vision, speech and generative-AI systems end to end, from research prototypes to production products.
Feb 2026 — Present
PelakGuard · pelakgard.ir
Building PelakGuard as an independent product — a live service that detects, masks and reads license plates to protect car owners' privacy.
Dec 2023 — Apr 2024
Nazru Company · Tehran, Iran
Built generative-model pipelines for object and shadow removal in satellite imagery, taking research prototypes toward production quality.
Sep 2022 — Feb 2024
Malek Ashtar University of Technology · Tehran, Iran
Research associate working on domain adaptation from simulation to real space for object detection in aerial images, with results published as preprints.
Jan 2020 — Sep 2020
Malek Ashtar University of Technology · Tehran, Iran
Assisted the graduate deep learning course: lab sessions, project supervision and evaluation.
Sep 2019 — Jan 2020
Malek Ashtar University of Technology · Tehran, Iran
Assisted the machine learning course: lab sessions, assignments and evaluation.
03 — Education
M.Sc. · Electrical Engineering
Malek Ashtar University of Technology · Tehran, Iran
Thesis: improving the accuracy of man-made object detection in aerial and satellite images using deep learning.
B.Sc. · Electrical Engineering
Bu-Ali Sina University · Iran
Final project: temperature control system based on an AVR microcontroller.
Diploma
Ayatollah Madani High School · Hamadan, Iran
04 — Skills
05 — Projects
A Windows dictation tool that turns Persian speech into text in any program with one global hotkey — fully offline on the user's machine, with the live transcription beside the cursor. Built on an offline speech-recognition model (Shenava-Koochik, 114M parameters, via sherpa-onnx) running in real time on CPU, with voice punctuation commands, spoken-number to digit conversion, and stability work that cut the final word error rate from 34.1% to 30% in real-speech tests.
A Persian text-to-speech tool that turns any text into natural speech offline on CPU and clones a new voice from a few seconds of reference audio. The model was optimized and exported to ONNX Runtime to run without PyTorch — cutting dependencies from ~1.2 GB to ~200 MB with output quality identical to the original engine — behind a Persian RTL interface with phoneme editing, built-in and user-uploaded voices, and long-text support. Model package and demo published on Hugging Face, and the project has 180+ stars on GitHub.
A live product that protects car owners' privacy when they publish vehicle ads: upload a photo, the plate is detected and masked (blur, redaction box, or custom logo), and the protected image is ready in seconds. Detection and recognition run on my own lightweight models — trained on a Persian plate dataset and optimized for real-time use — with four engines for standard, angled/rotated, OCR (Persian/English/Arabic) and hybrid cases, an interactive box editor, and a full Persian/English interface.
A question-answering assistant for Persian pharmacy reference content: ask about a drug in natural Persian — indications, dosage, brands, side effects — and get answers grounded in the source text with citations, running entirely locally so sensitive material never leaves the machine. Includes the full RAG pipeline (ingestion, chunking, Qwen3 embeddings via Ollama, semantic retrieval), a bilingual Persian/English interface, chat sessions, and a documents dashboard.
Deep-learning instance segmentation of cerebral hemorrhage regions in medical scans.
Research project
Generative inpainting pipeline that removes unwanted objects and their cast shadows from satellite imagery.
For: Nazru Company
Hybrid convolutional–transformer network recognizing emotions from facial imagery.
Personal interest
Detects missing tooth regions in panoramic radiographic images to support dental implant planning.
Academic research
Deep neural network detecting dental caries in radiographic images.
Academic research
Bridged simulation and real-world domains so detectors trained on synthetic data generalize to real aerial imagery.
For: Malek Ashtar University of Technology
Master's thesis on improving the detection accuracy of man-made objects in aerial and satellite images using deep learning methods — the foundation of my research direction in overhead imagery.
For: Malek Ashtar University of Technology
Automatic number-plate recognition pipeline for vehicles on highways.
Industry project
Automated reading of electricity and gas meters from images.
Industry project
Face recognition pipeline developed during my research assistantship.
For: Malek Ashtar University of Technology
07 — Credentials
AI for Medical Diagnosis
Coursera · DeepLearning.AI
AI for Medical Prognosis
Coursera · DeepLearning.AI
Convolutional Neural Networks in TensorFlow
Coursera · DeepLearning.AI
Introduction to TensorFlow for AI, ML & DL
Coursera · DeepLearning.AI
Rank 811
Master's entrance exam, among nationwide participants
Sep 2017
Rank 2600
Bachelor's entrance exam, among nationwide participants
Sep 2012
08 — Contact
Open to roles and collaborations in machine learning and computer vision — full-time, contract or research. The fastest way to reach me is email.
Ali Akbar Kiaei
Post-doctorate in Artificial Intelligence for Medicine
Sharif University of Technology
Additional references available on request.