Bench-marking Inference time scaling strategies on MLE-bench for measuring how well AI agents perform at machine learning engineering

Graduation Project

Active Learning experiments using Bayesian neural networks (BNNs)

No description.

automatic tree search llm based agent

Using Information theory to quantify uncertainty in deep learning models

The content of the Introductory sessions at IndabaX

Personal portfolio - Asim Osman

No description.

Online Goal-Conditioned Reinforcement Learning in JAX. ICLR 2025 Spotlight.

No description.

This project presents the development of a col- laborative filtering-based recommender system using Alternating Least Squares (ALS), with la- tent factor embeddings for users, movies along- side with their biases

Movie Recommender API - FastAPI + Collaborative Filtering

No description.

No description.

No description.

python implementation of SHA, AES and Vigenere Security algorithms

Neuromatch-Academy-Project-Code

No description.

No description.

Web Development projects

No description.

This repository contains an experimental implementation of **Fine-Grained Credit Assignment for RL Training (CAL)** on top of Google's Tunix framework. This research explores token-level reward assignment to improve training stability and sample efficiency in reinforcement learning for large language models.

Multi-Agent Reinforcement Learning with JAX

Materials and notes for the Science and Engineering of Large Language Models workshop at AIMS South Africa, Cape Town, 2025.

🦁 A research-friendly codebase for fast experimentation of multi-agent reinforcement learning in JAX

No description.

Efficient baselines for autocurricula in JAX.

The best ChatGPT that $100 can buy.

No description.

DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.

No description.

Composable transformations of Python+NumPy programs: differentiate, vectorize, JIT to GPU/TPU, and more

🏛️A research-friendly codebase for fast experimentation of single-agent reinforcement learning in JAX • End-to-End JAX RL

No description.

Thesis Essay

AIDE: the Machine Learning CodeGen Agent

The implementation incorporates various critical reinforcement learning techniques, including Q-learning, epsilon-greedy exploration, and neural network-based function approximation.

Course repository for Simulation and Inference for Neuroscience

This project demonstrates the step-by-step implementation of a neural network from scratch using Python, covering data preparation, core functions, training, and visualization.

This project explores binary classification methods using decision stumps and decision trees, focusing on their performance, overtraining tendencies

The challenge is to build a classifier that distinguishes electron showers (signal) from hadron showers (background) based on their depth and width.

This project demonstrates the application of the REINFORCE algorithm, a policy gradient method, to solve the LunarLander-v2 environment using reinforcement learning.

This project implements the REINFORCE algorithm to solve the CartPole-v1 environment. The algorithm uses policy gradients to optimize the agent’s performance by reinforcing actions that lead to higher returns.

No description.

This project aims to develop and evaluate a vehicle re-identification system using the VeRi dataset. Leveraging a pre-trained ResNet50 architecture and triplet loss, we designed a system to learn embeddings for vehicle images. Two triplet mining strategies—random and semi-hard

This project demonstrates the implementation of a fully convolutional network for image denoising using the color version of the LFWcrop dataset. It aims to reconstruct clean images from their noisy counterparts by leveraging a small yet effective autoencoder architecture.

training a classification model for CIFAR-10, but this time with transfer learning, inestigation model focus with sailancy maps

This project explores the principles of active learning through the Forest Covertype Dataset, using various query strategies to demonstrate its impact on model performance and data efficiency.

Implementing and evaluating various Convolutional Neural Network (CNN) architectures to classify images from the CIFAR-10 dataset. Through systematic experimentation, we explore the effects of model complexity, regularization techniques, and data augmentation on classification accuracy.

No description.

No description.

Using satellite imagery and computer vision, farmers can map fields, track crop health, and receive actionable recommendations via a user-friendly dashboard

Project on crop monitoring-Amundata

Using satellite imagery and computer vision, farmers can map fields, track crop health, and receive actionable recommendations via a user-friendly dashboard.

practice with python on competitive programming