About

ML Software Engineer with 3+ years of experience designing and deploying scalable software systems across Computer Vision, LLMs/RAG, and Speech Processing. With a track record of building industry-driven solutions, from real-time inference pipelines and cloud-deployed ML services to safety-critical and diagnostic systems.

Bridging research-grade methods with robust, production-ready software design. I hold an M.Sc. in Computer Engineering from Politecnico di Torino, with a thesis on voice conversion conducted at Technische Universitat Darmstadt. Driven by a strong interest in systems and software architecture, I am looking to transition toward large-scale design roles in the high-tech industry.

Projects

Experience

Machine Learning Engineer

Nov 2024 - Present
Zirak Turin
  • Architected a natural-language-to-SQL platform for Poste Italiane's logistics division (FastAPI + React/TypeScript + LangGraph), featuring Ollama for local LLM support, multi-database support (PostgreSQL, MySQL, MongoDB), and auto-detected visualizations, enabling non-technical operators to query sorting data in Italian without SQL knowledge.
  • Led a team of two to build a cross-platform LLM meeting assistant (Electron/React + Python + FastAPI) with RAG (LangChain/LangGraph, Milvus), integrating speaker verification, and real-time transcription on AWS Bedrock for meeting summaries and Q&A.
  • Built a rail-safety Computer Vision pipeline (YOLOv8/PyTorch) to detect sign defects and segment/track rails from nadir and frontal views.
  • Defined user stories, architecture designs, and detailed design requirements for safety and diagnostics modules based on customer needs, ensuring alignment with project specifications and driving an efficient architecture.
  • Designed and implemented a custom U-Net for multispectral imagery, automated data ingestion, training and evaluation with Airflow, outperforming internal baselines (Random Forest, XGBoost, LightGBM) in F1 by ~10%.
  • Optimized an aquatic-vegetation detection pipeline on AWS using multispectral satellite data, keeping accuracy drift <3% and reducing manual image review by ~80%.
  • Deployed an LLM assistant for consultancy and HR workflows, serving HuggingFace models via vLLM/TorchServe with MLflow tracking and Grafana dashboards, cutting manual document processing by ~50 hours/month and enabling the product for other clients.

Visiting Researcher - Master Thesis

Sep 2023 - Jul 2024
Technische Universitat Darmstadt | CROSSING Darmstadt
  • Designed a Diffusion Transformer for voice conversion (Python, PyTorch) on HPC GPUs, improving speaker similarity by ~9% over open-source baselines with fewer sampling steps, enabling real-time conversion.

AI Engineer

Feb 2020 - Mar 2022
Part AI Research Center Tehran
  • Refined and fine-tuned an on-device keyword-spotting Android app in Java, using hard negative samples to reduce false positives by ~30%.
  • Optimized a speaker verification system on TensorFlow, fine-tuned on new domains that reduced the Equal Error Rate (EER) by ~2%.
  • Engineered Test-Driven Development TDD for Persian ASR and NLP services; wrote unit/integration tests to reach 100% coverage and set up Jenkins CI/CD to automate builds, tests, and releases.
  • Mentored new team members on clean, production-ready Python (structure, testing, logging), improving code quality and safety.
  • Implemented Neural Network architectures for scalable products, optimizing for both cost-efficiency and operational safety.

Associate Software Developer - Internship

Jan 2023 - Jul 2023
AROL S.p.a Canelli
  • Developed a synthetic machinery data generator (TensorFlow + Flask + MongoDB) and a customer testing dashboard, packaged with Docker Compose (client/server/DB) and REST API docs to avoid exposing proprietary data.
  • Designed KPIs and Python tests to validate synthetic vs. real machinery signals and API responses, integrating these checks into the team's CI pipeline and reducing manual QA effort for customer demos.
  • Reviewed over 5,000 lines of production Python and ML pipeline code (training, inference, data preprocessing), adding tests and reducing compute cost while making models easier to debug and extend.

Computer Vision - Internship

Sep 2019 - Feb 2020
Megamouj Co. | University of Science and Technology (IUST) Tehran
  • Automated a traffic-flow monitoring system (C++/Python, OpenCV, YOLOv3) for vehicle detection, tracking, and flow estimation, reducing manual video review in a national-scale project.
  • Engineered a data analysis pipeline using NumPy and Pandas to structure annotation datasets, utilizing scikit-learn to compute performance metrics and validate system reliability.
  • Labeled and augmented traffic video data and trained/evaluated detection and tracking models in Python, tuning YOLOv3 hyperparameters and metrics to keep counting errors within acceptable limits across varying lighting and weather conditions.

Skills

Software Development Python, C/C++, SQL, Bash, Linux, Git, CI/CD, FastAPI, Grafana, Data Structures and Algorithms
Machine Learning NumPy, Pandas, scikit-learn, XGBoost, LightGBM, PyTorch, TensorFlow, MLflow, Airflow, GANs, U-Net
Large Language Models Transformers, RAG, LangChain, LangGraph, vLLM, TorchServe, Milvus, HuggingFace, Ollama
Cloud Technologies AWS (Bedrock, ECS, S3), Docker, Docker Compose
Reinforcement Learning Stable Baselines3, Optuna, MobileNetV2, YOLO

Education

M.Sc. Computer Engineering - Artificial Intelligence & Data Analytics

2022 - 2025

Politecnico di Torino, Turin

Courses: Advanced Machine Learning, System and Device Programming, Software Engineering, Web Applications

Thesis: RobVC: Robust Zero-Shot Self Supervised Voice Conversion (Funded by CROSSING Co.)

B.Sc. Electrical and Computer Engineering - Telecommunications

2016 - 2021

Iran University of Science and Technology, Tehran

Courses: Programming, Digital Signal Processing, Linear Algebra, Linear Control Systems, Communications

Thesis: Deep Learning-Based Vehicle Detection and Traffic Flow Analysis (Funded by Megamouj Co.)

Blog

View all posts

Contact

Feel free to reach out for collaborations or opportunities.

Email: amadreza.farahani@outlook.com

LinkedIn: linkedin.com/in/amadrezafrh

Location: Turin, Italy