cv

M.Sc. in Artificial Intelligence

Addis Ababa University, Addis Ababa, Ethiopia

• Thesis - RL based Dynamic Transformer Networks for efficient inference.

B.Sc. in Software Engineering

Addis Ababa University, Addis Ababa, Ethiopia

• Thesis - Dynamic Transformer Networks for faster inference.

Machine Learning Engineer

BitzBrain Research

• Reproduced papers and conducted NLP research on low-resource multilingual datasets.
• Implemented image and text retrieval systems with a multimodal language model, and vector databases.
• Significantly reduced training and deploying cost of language models through innovative techniques.

Software Engineer

Lumos Technologies

• Lead a development team
• Design System Architecture

Researcher & Lecturer

Addis Ababa University

• Teach courses - Intro to AI, Intro to Algorithm and Data Structure.
• Develop software products for the university
• Conduct AI/ML Research

Founder & Developer

Sefere App

• Launched a startup intending to boost the visibility of small businesses through mobile and web apps.
• Implemented a self-improving content-based recommendation and feed ranking system using various ML algorithms.

Computer Vision and Robotics Engineer

4Kbots AI and Robotics Lab

• Built a ROS-based RoboCup Platform
• Implemented SLAM and Object Detection algorithms

DynamicViT - Making Vision Transformer faster by skipping layers - GitHub

Addis Ababa University

• Developed a transformer network that employs a reinforcement learning (RL) agent to learn to skip irrelevant layers during processing an input, resulting in a 40% increase in throughput compared to the vanilla ViT

DistillEmb - Distilling word-embeddings via contrastive learning - GitHub

Addis Ababa University

• Compressed word embeddings into a CNN to address out-of-vocabulary issues, conserve memory, and facilitate cross-lingual transfer, resulting in a 7% improvement in F1 accuracy and 1400% less memory usage, as well as out-of-the-box low resource cross-lingual transfer capabilities.

Boosting image retrieval with accurately labeled small data - GitHub

Addis Ababa University

• Explored the utilization of a detailed labeled dataset to enhance the feature set in an image retrieval task, enabling a small, shallow network to surpass the performance of a more complex, larger model

Morphological generation for Wolayita using CNN based Seq2Seq model

Addis Ababa University

• Constructed a CNN-based encoder-decoder morphological analyzer for Wolayita language utilizing fewer parameters than LSTM-based models

Morphology-rich alphasyllabary embeddings - GitHub

Addis Ababa University

• Examined the relationship between language morphology and corresponding word embeddings, proposing a technique to enhance the syntactic features of embeddings

Urban Land Usage Optimization Using NSGA II - Jupyter Notebook

ROS and Gazebo-based Delta-3 robot model - GitHub

Genetic Algorithm based classroom timetable generator - GitHub

ICCV Student Volunteer

International Conference on Computer Vision (ICCV) 2023 - Paris, France

ICLR Student Volunteer

International Conference on Learning Representation (ICLR) 2023 - Kigali, Rwanda

Program Chair

Deep Learning IndabaX ASTU Conference - Adama, Ethiopia

Ambassador of Ethiopia

Zindi Africa - Addis Ababa, Ethiopia

General Co-Chair

AI in Ethiopia Conference - Addis Ababa, Ethiopia

Tutorial Chair

Deep Learning IndabaX WSU Conference - Wolaita Sodo, Ethiopia