Arman Ahangari Kiasari

AI Researcher

I apply AI to real-world problems, evaluate model performance through experimentation, and share findings through scientific publications, aiming to produce meaningful results.

Latest activity

About

I build and evaluate AI systems.

Download CV

Background

I am an AI researcher applying machine learning to structured data and deep learning to image analysis. I hold an MSc in Computer Engineering, specializing in Artificial Intelligence and Robotics, from Islamic Azad University, Central Tehran Branch. I have also worked on signal processing and algorithm development for sensor-based systems.

Focus

My research interests center on developing AI methods suited to the challenges of real-world data analysis. I am currently exploring generative adversarial networks and reinforcement learning, while broadening my understanding of signal and audio data. Looking ahead, I aim to investigate time-series modeling and sequence-learning approaches, particularly for video analysis, where meaningful patterns may emerge across successive frames. I am interested in how these methods can be adapted to different types of data and the specific problems each one presents.

Experience

Research groups I have worked with

  1. AI Research Engineer – Signal Processing

    Sep 2022 – Feb 2024

    Metable

    • Reviewed relevant literature to select suitable AI methods for signal analysis and detection.
    • Implemented selected methods in code and applied them to signal data collected through the Patriiot-01 platform.
    • Evaluated and tuned models to improve predictive performance and identify approaches suitable for real-time signal processing.
  2. Software Engineer – Monitoring Devices

    Aug 2019 – Nov 2021

    Roshan Andishan Radman

    • Developed signal-processing algorithms to filter sensor data and extract features from raw sensor signals.
    • Implemented estimation routines to calculate target measurements from sensor readings.
    • Refined and tested the algorithms under varying measurement conditions to improve the consistency of their estimates.
Work

Papers and projects

Published IEEE · IBAAIDS 2026 ·

Improving Gallstone Disease Prediction through Composite Feature Optimization with CatBoost

On the public UCI Gallstone dataset, which holds clinical, bioimpedance and laboratory measurements of 319 individuals from a hospital clinic, gallstone disease is predicted early without any imaging. Features are ranked with ANOVA and merged into composite features, and a CatBoost classifier is trained on the optimized set. It reaches 89% accuracy and 0.94 AUC.

In preparation Last update:

A Hybrid Convolutional Neural Network for Pediatric Congenital Heart Disease Classification from Chest X-rays

On the public Kaggle congenital heart disease collection of pediatric chest X-rays, three defect types (ASD, VSD and PDA) are separated from normal hearts to support early diagnosis. A hybrid ResNet50–DenseNet121 convolutional framework is trained on the images. On the test set it reaches 82% accuracy and 0.91–0.95 ROC-AUC.

With Parvaneh Asghari

Ongoing Last update:

TC-LFM: Triplet-Consistent Latent Flow Matching for Conditional Generation of Cardiopulmonary Sounds

On HLS-CMDS, the public UCI collection of 535 heart and lung recordings made with a digital stethoscope on a clinical manikin, a model is being developed to generate class-conditioned heart, lung and mixed sounds that could enrich training data for cardiopulmonary diagnosis. It combines a mel-spectrogram VAE, triplet alignment and conditional flow matching. The work is at an early stage and results are not yet available.

Skills

Tools and languages I work with

Artificial Intelligence & Machine Learning Python

  • PyTorch
  • Torchvision
  • scikit-learn
  • OpenCV

Data Science & Scientific Computing Python

  • NumPy
  • Pandas
  • SciPy
  • Matplotlib
  • Seaborn
  • Plotly

Research & Development Tools

  • VS Code
  • Jupyter Notebook
  • Google Colab
  • Git
  • GitHub
  • TensorBoard
  • SQL
  • draw.io
  • ResearchRabbit
  • Mendeley
  • Microsoft Word
Contact

Get in touch