CV

Saeedreza Zouashkiani

saeedzou2012@gmail.com
+98-915-415-76-30
Mashad, Iran, IR

Summary

AI Researcher with expertise in Speech Processing, Large Language Models, and Reinforcement Learning. Currently at the Foundation AI Models Lab, Sharif University of Technology.

Education

  • M.Sc. in Communication Systems
    2024-06
    Sharif University of Technology
    GPA: 3.7/4.0
    Courses: Deep Learning, Convex Optimization, Adaptive Filters
  • B.Sc. in Electrical Engineering
    2021-09
    Amirkabir University of Technology
    GPA: 3.2/4.0
    Courses: Statistical Learning, Linear Algebra

Work Experience

  • AI Researcher, Automatic Speech Recognition
    2024-08 - Present
    Foundation AI Models Lab, Sharif University of Technology
    Conducted research on ASR architectures, trained top-performing models, and led a project integrating continual dataset growth, achieving state-of-the-art results for Persian ASR.
  • AI Consultant
    2025-05 - Present
    Bimeh Dot Com, Sepas Holding
    Developed automated pipelines for document understanding, researched RL applications in finance, and built an OCR pipeline for car insurance card processing.
  • AI Researcher, Text-to-Speech
    2025-03 - Present
    Foundation AI Models Lab, Sharif University of Technology
    Developed a G2P pipeline for Persian, compiled a high-quality audiobook corpus, and trained the first bilingual Persian-English zero-shot TTS model.
  • AI Researcher, Voice Conversion
    2025-03 - Present
    Foundation AI Models Lab, Sharif University of Technology
    Leading research on voice conversion, including emotion transferability, and fine-tuning a model on 1,000 hours of Persian audio data.
  • AI Researcher, Speech Data Creation Pipeline
    2024-08 - 2025-07
    Foundation AI Models Lab, Sharif University of Technology
    Collected and processed over 11.8k hours of audio data, created a scalable validation pipeline, and produced large-scale datasets for ASR, TTS, and VC.
  • Graduate Researcher, MSc Thesis
    2021-09 - 2024-06
    Sharif University of Technology
    Proposed a hybrid meta-reinforcement learning framework for dynamic edge offloading, reducing trainable parameters and leveraging deep learning for scalable edge orchestration.

Skills

Programming

  • Python
  • Vibe Coding
  • MATLAB
  • Bash

AI Frameworks

  • PyTorch
  • TensorFlow
  • Scikit-learn
  • OpenCV
  • Hugging Face

AI Domains

  • Speech Processing
  • Natural Language Processing
  • Computer Vision
  • Reinforcement Learning
  • Prompt Engineering

Simulation & Tools

  • Git
  • Docker
  • Linux (Ubuntu)

Publications

  • PersianVox: A Prosody-Aware Approach for Speech Dataset Generation from In-the-Wild Data
    2026
    arXiv:2609.19324
    Zouashkiani, S., Khalesi, S., Soleimani Roudi, S., Amini, S., and Ghaemmaghami, S.
  • A Survey on Non-Intrusive ASR Refinement: From Output-Level Correction to Full-Model Distillation
    2025
    arXiv:2508.07285
    Peyghan, M. R., Rajabi, F., Soleimani Roudi, S., Zouashkiani, S., Amini, S., and Ghaemmaghami, S.

Teaching

  • Deep Learning
    2024
    Sharif University of Technology
    Role: Head Teaching Assistant
  • Introduction to Machine Learning
    2023
    Sharif University of Technology
    Role: Teaching Assistant
  • Machine Learning
    2023
    Sharif University of Technology
    Role: Course Coordinator
  • Communication Systems
    2023
    Sharif University of Technology
    Role: Teaching Assistant
  • Adaptive Filter
    2022
    Sharif University of Technology
    Role: Teaching Assistant
  • Linear Algebra
    2021
    Amirkabir University of Technology
    Role: Guest Lecturer

Portfolio

  • Persian ASR, Translation & Summarization App
    2025
    Gradio
    Developed a Gradio app for Persian ASR, with support for multilingual translation and summarization of transcriptions using Gemini API.
  • ParsNorm: Advanced Persian Text Normalization
    2024
    Github
    Developed a comprehensive text normalization pipeline and a unique English-to-Persian transliteration tool.
  • Pin Insulator Instance Segmentation with YOLOv8
    2023
    Github
    Trained a YOLOv8 segmentation model on the CLIPD dataset, achieving 96.41% mAP50.

Languages

  • Persian
    Native
  • English
    Fluent

Interests

  • Speech Processing
    Automatic Speech Recognition (ASR), Text-to-Speech (TTS), Voice Conversion (VC)
  • Large Language Models
    Retrieval-Augmented Generation (RAG), Agentic Frameworks
  • Reinforcement Learning
    Optimization Techniques, Exploration Strategies