Firas Sawab

AI Engineer working on Machine Learning, Deep Learning, and Agentic Systems and Engineering.

View work GitHub

From model to product.

I’m an AI engineer. I train deep learning models, and I build the software and agents around them so people can actually use them. I’m finishing my degree in Artificial Intelligence and Computer Science.

Models, retrieval, agents.

I train and evaluate vision and language models in PyTorch, ground LLMs in private data with retrieval, and build tool-using agents that plan their steps and check in before anything risky.

  • Python
  • PyTorch
  • Transformers
  • LLMs
  • RAG
  • Ollama
  • ChromaDB
  • OpenCV
  • MediaPipe
  • NumPy
  • pandas
  • scikit-learn

Here’s what I’ve built.

Projects

Tell Jay what you want done and it plans, browses in its own browser, reads and writes files, builds and previews code, and reports back while you watch it work live. A Tauri desktop app with a Rust core and a SolidJS interface, driven by a TypeScript agent that connects to MCP servers, asks before risky actions, runs scheduled routines and takes tasks over Discord.

A student platform built around Hermes, an AI coach that turns a student’s courses, goals and progress into an adaptive learning roadmap, quizzes and slides, and proposes changes for approval instead of doing the work for them. It also covers group projects, career matching and a private inbox sorted by a local email classifier. React and Vite on the front end, FastAPI and SQLite behind it.

View code on GitHub (opens in a new tab)

Thousands of movies are embedded by what they’re about and projected into 3D, so similar films sit close together. Search a title to fly to it and its nearest neighbours, or ask the assistant, a tool-calling chatbot that searches the movie database and the web through MCP servers. Built with a team for a machine learning course.

View code on GitHub (opens in a new tab)

A local, privacy-first retrieval-augmented generation system. Documents are chunked, embedded with Sentence Transformers and stored in ChromaDB; each question pulls the most relevant passages by semantic search and hands them to a local LLM through Ollama.

View code on GitHub (opens in a new tab)

MediaPipe extracts 63 hand-landmark coordinates per frame, and a lightweight PyTorch classifier reads them in real time through OpenCV. Includes a data-collection script and a live demo on Hugging Face.

View code on GitHub (opens in a new tab) Try the live demo (opens in a new tab)

A transformer encoder in PyTorch, with multi-head self-attention and positional encoding, trained for sentiment classification.

View code on GitHub (opens in a new tab)

Forward pass, loss functions, optimizers and backpropagation implemented directly in NumPy, without an autograd library.

View code on GitHub (opens in a new tab)

A multilayer perceptron that learns to map inputs to outputs for mathematical functions and whole images, with a live plot that redraws every epoch.

View code on GitHub (opens in a new tab)

A convolutional network for text classification based on the CNN-for-sentence-classification paper, comparing kernel sizes (effectively n-gram widths) and pooling strategies.

View code on GitHub (opens in a new tab)

A collection of convolutional architectures implemented in PyTorch for image-classification experiments, trained and benchmarked under the same setup.

View code on GitHub (opens in a new tab)

Coursework

DeepLearning.AI

  • Neural Networks and Deep Learning
  • Improving Deep Neural Networks: Hyperparameter Tuning, Regularization and Optimization
  • Structuring Machine Learning Projects
  • Convolutional Neural Networks
  • Sequence Models

IBM

  • Exploratory Data Analysis for Machine Learning

University of Michigan

  • Applied Machine Learning in Python
  • Introduction to Data Science in Python
  • Data Collection and Processing with Python
  • Python Classes and Inheritance
  • Python Functions, Files, and Dictionaries
  • Python Basics

Contact

Open to AI engineering roles in machine learning, deep learning, LLM applications and agentic systems.

FirasSawab@outlook.com