Iheb Gafsi

Machine Learning Engineer | Mathematics & Physics Enthusiast

I study learning and computation in large-scale neural systems using tools from statistical physics, dynamical systems, information theory, and geometry. I also build physics‑informed and scientific ML systems for modeling complex physical phenomena.

12,076 Followers
1,900+ Professionals Coached
45+ Research Projects
Portrait

About Me

My research lies at the intersection of statistical physics, dynamical systems theory, and machine learning. I investigate the fundamental mechanisms underlying learning and computation in large-scale neural systems through the lens of statistical mechanics, information geometry, and optimization theory. My work addresses questions of generalization, scaling laws, and the emergence of structure in high-dimensional representations.

On the applied side, I develop physics-informed neural networks and scientific computing frameworks for modeling complex physical phenomena—from quantum systems to fluid dynamics. This involves numerical methods for PDEs, uncertainty quantification, and interpretable models that respect physical constraints. I also work on accelerated AI systems and distributed ML pipelines, with particular focus on efficient training dynamics and memory-constrained optimization.

Based in Tunisia

Physics & Mathematics

Statistical Mechanics Quantum Mechanics Dynamical Systems Information Theory Differential Geometry Stochastic Processes Optimization Theory Numerical Analysis

Scientific Computing & ML

Physics-Informed Neural Networks PDE Solvers Bayesian Inference Monte Carlo Methods Large Language Models Computer Vision Quantum Computing Distributed Systems

Research Interests

Physics of Learning

  • Statistical mechanics of generalization and scaling
  • Dynamical systems and optimization landscapes
  • Information theory and representation geometry

Scientific Computing

  • Numerical methods for PDEs and inverse problems
  • Physics‑informed neural networks (PINNs, KANs)
  • Uncertainty quantification and Bayesian inference

Quantum & Complex Systems

  • Quantum information and quantum‑ML
  • Complex networks and stochastic processes
  • Interpretable models for physical systems
Variational inference Bayesian neural nets PDE‑constrained optimization Monte Carlo & MCMC Langevin dynamics Spectral methods Finite‑element/volume Information geometry

Professional Experience

Technical Course Writer

DataCamp

Jan 2025 – Present New York, United States (Remote)

Creating educational materials on Artificial Intelligence and Machine Learning, focusing on simplifying complex technical concepts for learners worldwide.

Machine Learning AI Education Multi-agent Systems

Deep Learning Instructor

NVIDIA

May 2024 – Apr 2025 Stuttgart, Tunisia (Hybrid)

Coached 1,900+ professionals and collaborated with universities. Delivered workshops on Accelerated AI using NVIDIA's NeMo Services, Cloud, RAPIDS, and Jetson Nano.

  • Efficient Large Language Models Customization
  • Computer Vision for Industrial Inspection
  • Building Transformer-Based NLP
  • Data Parallelism for Multi-GPU Training
Deep Learning NVIDIA Technologies Training & Education

AI Research Scientist

WiseVision AI Technologies

Oct 2024 – Feb 2025

Advanced research in predictive modeling, multi-agent systems, and computer vision applications.

Predictive Modeling NER Intent Classification

Associate Machine Learning Researcher

Callem AI

Apr 2024 – Nov 2024 Île-de-France, France (Hybrid)

Automated telephony with energy-efficient Conversational AI on distributed systems.

Production Work:
  • Developed Graph Data for Telephony Conversational AI
  • Architected Multi-Agent Systems and published MASAS paper
  • Implemented Multi-Turn Context Response Selection pipeline
  • Published NoRA paper (Neuro-Evolution of Low Rank Adaptation)
Research Work:
  • Implemented Neural Radiance Fields (NeRF) for 3D-vision
  • Fine-tuned Whisper-based TTS and ASR models
Multi-agent Systems Distributed Systems GraphQL

Technical Advisor

Revenger AI

Sep 2024 – Oct 2024 Cologne, Tunisia (On-site)

Employed advanced AI for reverse engineering tasks, including vulnerability identification and data extraction. Developed AI-driven reverse engineering frameworks for efficiency and precision.

Multi-agent Systems Data Engineering Automated Feature Engineering

Data Specialist

Mindrift

Jul 2024 – Aug 2024 Lund, Sweden (Remote)

Designed anomaly detection pipelines and ensured data reliability through cleansing and correction. Documented data quality findings and recommendations.

Big Data Analytics Data Engineering Anomaly Detection

Mentor

Teens in AI

Mar 2024 – May 2024 United Kingdom (Remote)

Mentored teens in AI, code, and AI ethics for the Global Techathon, developing a Healthcare Information Technology (HIT) solution.

RAG AI Ethics Healthcare IT Mentoring

GenAI Engineer

Silver Brain AI AG

Feb 2024 – May 2024 Switzerland (Remote)

Advanced LLM development and optimization for chemistry applications.

  • Trained Mistral 7B-Instruct-v0.2 on chemistry concepts, outperforming GPT-4 on 8 tasks
  • Improved LLM performance from 68% to 92% using QLoRA on Azure VM
  • Aligned LLMs to eliminate hallucination and limited action scope
  • Developed synthetic data generation systems, increasing result diversity by 47%
  • Built auto-evaluation tools for LLM-based solutions
Azure ML LLM Fine-tuning QLoRA Synthetic Data

Data Engineer Intern

Silver Brain AI AG

Dec 2023 – Feb 2024 Zurich, Switzerland

Foundation work in data engineering and ML systems for chemistry applications.

  • Designed ETL pipelines for chemistry data and utilized Apache Cassandra for scalable storage
  • Implemented attention-based OCR systems and generated synthetic data for benchmarking
  • Scraped licensed chemistry data using Puppeteer
ETL Pipelines Apache Cassandra OCR Systems MLOps

MLOps Engineer

PeddyMark Ltd. / Animal Tracker

Sep 2023 – Nov 2023 Redhill, England (Remote)

Developed ML models for pet tracking using geospatial data and SAR applications. Analyzed pet location data and incorporated environmental factors for predictive modeling.

  • Created geofencing and activity recognition models for pet behavior insights
  • Implemented search and rescue (SAR) optimization algorithms
Geospatial ML MLOps Predictive Modeling

Leadership & Volunteering

AI Training Manager

ACM INSAT Student Chapter

Sep 2024 – Present

Leading AI training initiatives and educational programs for computer science students.

Writer

INSAT Press

Sep 2024 – Present

Contributing articles on human rights and technology topics for the institutional publication.

Human Rights

Head of Machine Learning

Google Developer Student Club - INSAT

Sep 2024 – Present

Leading ML initiatives and organizing workshops for the developer community.

Chairman

Data Overflow

Nov 2023 – Sep 2024

Led data science community initiatives and organized technical events.

Member

IEEE Computer Society

Sep 2023 – Present

Active participation in professional development and technical conferences.

Core Team Member

Google Developer Student Club - INSAT

Aug 2021 – Present

Long-term contributor to science and technology initiatives in the student community.

Science & Technology

Additional Learning

Cosmology

MIT OpenCourseWare

8.942

Advanced study of the universe's structure, evolution, and fundamental physics.

Quantum Computation

MIT OpenCourseWare

18.435J

Theoretical foundations and practical applications of quantum computing systems.

Quantum Theory II

MIT OpenCourseWare

8.322

Advanced quantum mechanics and its applications in modern physics.

Education

INSAT - Institut National des Sciences Appliquées et de Technologie

Engineering, Applied Sciences and Technology Aug 2022 – Jun 2027
Activities: ACM, IEEE, GDSC, INSAT Press

Mathematics

Stochastic Processes, Statistical Learning, Convex Optimization, Advanced Statistics, Operational Research, Numerical Methods

Computer Science

Computer Architectures Engineering, Operating Systems Engineering, Smart Devices, Embedded Intelligent Systems and IoT

Networks & Telecommunications

Discrete Time Signal Processing, Protocol Engineering, Random Signal Processing, 5G Networks

Lycée de Jemmel

Baccalaureate, Computer Science Jun 2021 – Jun 2022
4th Top Student at National Level - 11th nationally

Received municipal, provincial, and ministerial honors. Authored Python course for scientific computing and created chemistry handbook.

Featured Projects

Gaussian B-Splined Kolmogorov Arnold Physics Informed Neural Networks

Jul 2025

Utilized Gaussian B-splines in Kolmogorov-Arnold approximators for interpretable ML. Demonstrated KANs' universality for solving parametric differential equations.

Machine Learning Scientific Computing Fluid Mechanics
View on GitHub

Voice Controlled Robotic Vehicle

Jun 2025

Developed a voice-controlled robotic vehicle using LPU-based Whisper Large for millisecond response times.

Robotics Speech Recognition Real-time Systems

Optimizing Restricted Boltzmann Machines via Quantum Annealing

Jan 2025 – May 2025

Proposed a quantum restricted Boltzmann machine (QBM) for accurate data representations in high-dimensional Hilbert spaces.

Quantum Computing Statistical Physics Machine Learning

Kolmogorov Arnold Neural Networks Framework

Jan 2025

Python implementation based on the Kolmogorov-Arnold Representation Theorem for regression and classification tasks.

Neural Networks Python Framework Development
View on GitHub

Unstable Diffusion

Dec 2023 – Jan 2024

Built a generative model from scratch using PyTorch, based on Stable Diffusion concepts.

Generative AI PyTorch Computer Vision
View on GitHub

NEAT Framework

Jun 2023 – Jul 2023

Implemented a Python framework for Neuro Evolution of Augmented Topologies.

Evolutionary Algorithms Reinforcement Learning Neural Networks
View on GitHub

Credit Card Fraud Detection MLOps

Nov 2023

Implemented fraud detection with MLFlow and ZenML for streamlined ML lifecycle management and automated model deployment.

MLOps MLFlow ZenML Fraud Detection

Graph Theory Odyssey Book

Aug 2023 – Sep 2023

Authored a comprehensive book covering advanced graph theory topics, algorithms, and applications in computer science.

Graph Theory Algorithms Technical Writing
View on GitHub

NLP Projects Repository

Jun 2023 – Jul 2023

Collection of natural language processing projects using various techniques and modern frameworks including TensorFlow and PyTorch.

NLP TensorFlow PyTorch Text Processing
View on GitHub

Python Full Course

Aug 2021 – Nov 2021

Comprehensive documentation of Python libraries for data scientists and analysts, covering essential tools and techniques.

Python Data Science Education Documentation
View on GitHub

SMART Pet Plant

Associated with INSAT

IoT project using ESP32/Arduino to monitor and display plant conditions via sensors, creating an interactive pet plant experience.

IoT ESP32 Arduino Sensors
View on GitHub

Horror Game with Unreal Engine

Jun 2019 – Jan 2020

Developed a horror game using Unreal Engine and C++ with intelligent zombie AI powered by decision trees for realistic gameplay.

Unreal Engine 4 C++ Game AI Decision Trees

Publications & Research

RAYGo: Retrieve As You Go for Large-Scale Asynchronous Action Planning

Zenodo, Sep 1, 2024

Introduced a Chain-of-Actions and Retrieve-As-You-Go approach for LLM action planning, outperforming GPT-4 by 29%.

Linear-Time Sequence Modeling with Selective State Space Mamba

Zenodo, Jan 15, 2023

Presented Mamba, a linear-time architecture outperforming Transformers for long sequences.

Transformers

Zenodo, Aug 13, 2023

Overview of the Transformer architecture, focusing on self-attention and scalability in NLP.

Feed Forward Neural Networks

Zenodo, Jul 9, 2023

Detailed the architecture and applications of FFNNs as universal approximators.

Honors & Awards

First Place

National Robotics Weekend

Jun 2025

Won for developing an astronaut robot for Mars sample collection.

Google Developer Nanodegree

Google

Feb 2020

One Million Arab Coders Contest Winner

Ministry of AI, UAE

Feb 2020

Get In Touch

Let's Connect

Interested in collaborating on AI research, discussing machine learning projects, or exploring opportunities in accelerated AI systems? I'd love to hear from you.

Tunisia
Available on LinkedIn
12,076 Followers

Languages

Arabic (Native) English (Professional) French (Professional) German (Working)