About Me
LLM reasoning, grounding, robustness, and evaluation using multilingual and cultural variation as a scientific lens.
I am currently a PhD student and research assistant at Saarland University supervised by Prof. Dr.-Ing. Philipp Slusallek and Prof. Dr. Dietrich Klakow. With over four years of experience in the machine learning industry, I have worked in diverse applied research roles, including Applied Research Data Scientist at Iquartic and Applied ML Developer at SingularityNet/iCog-Labs.
I hold an MSc in Mathematical Sciences – Machine Intelligence from AIMS-AMMI and a Bachelor of Science in Software Engineering from Addis Ababa Institute of Technology. I am an active member of the Masakhane, ETHIO NLP and AI-Grid community.
My research focuses on effective domain adaptation and evaluation methods for large language models (LLMs) in low-resource human and data languages. The overarching goal of my work is to advance the development of AI systems that are accurate, interpretable, robust, and culturally grounded, while remaining human-centered and contextually aware. I investigates how LLMs can better understand, reason, and communicate across linguistic, cultural, and structural boundaries, bridging the gap between human meaning and machine representation. Ultimately, my work aims to create socially aware and context-sensitive AI systems that perform reliably across diverse languages and data modalities.
Education
Universität des Saarlandes, Saarbrücken, Germany
PhD candidate
September 2022 – December 2026 [Expected]
- Thesis: Understanding, Confusion, and Consequence of LLMs in Low Resource Languages.
- Supervisor: Prof. Dr.-Ing. Philipp Slusallek and Prof. Dr. Dietrich Klakow.
African Institute for Mathematical Sciences, Accra, Ghana
Master of Science, Mathematical Sciences – Machine Intelligence
October 2019 – March 2021
- Thesis: Exploring Data Imbalance and Modality Bias in Hateful Memes.
- Supervisor: Prof. Dr. Marcus Rohrbach.
Addis Ababa University, Addis Ababa, Ethiopia
Bachelor of Science, Software Engineering
September 2012 – December 2017
- Project: Amharic online handwriting recognition keyboard.
- Supervisor: Dr. Natnael Argaw Wondimu
- Recognition: Certification Of Recognition For Outstanding Graduation Project.
Experience
University of South Florida, Tampa, Florida
PhD Internship
July 2025 – Present
- Socio-Cultural Large Language Models and Low Resource Languages.
- Guest Lecture: University of South Florida.
Universität des Saarlandes, Saarbrücken, Germany
Research Assistant | HPC cluster support
September 2022 – Present
- Conducting research in NLP, Multimodal learning, and AI/ML.
- Managing HPC clusters and multi-GPU model training.
- Providing HPC-related training and technical support.
iQuartic, New Wave, Maryland, USA
Applied Research Data Scientist
March 2021 – October 2022
- Developed AI solutions for automated risk adjustment, transitioning pipelines to in-house solutions.
- Built end-to-end pipelines for medical OCR, spell checking, and ICD code extraction.
- Managed data collection, model building, deployment, and testing phases.
SingularityNet, Icog-Labs, Addis Ababa, Ethiopia
Applied ML Developer
June 2017 – August 2022
- Developed CNN, GAN, and VAE models for fake news and plant disease detection (97–98% accuracy).
- Implemented web recommendation systems using Doc2Vec and DeepCTR.
- Built large-scale image retrieval systems and conducted visual graph analytics.
Icog-Labs Software Consultancy, Addis Ababa, Ethiopia
Internship - Junior Programmer
Jan 2015 – May 2017
- Developed (YANETU), an AI-powered teaching assistant for children.
- Built product components using Unity3D, Blender, and OpenCog NLP.
Publications
-
AfrIFact: Cultural Information Retrieval, Evidence Extraction and Fact Checking for African Languages
April 2026
Assessing the veracity of a claim made online is a complex and important task with real-world implications. When these claims are directed at communities with limited access to information and the content concerns issues such as healthcare and culture, the consequences intensify, especially in low-resource languages. In this work, we introduce AfrIFact, a dataset that covers the necessary steps for automatic fact-checking (i.e., information retrieval, evidence extraction, and fact checking), in ten African languages and English. Our evaluation results show that even the best embedding model... -
Ethio-ASR: Joint Multilingual Speech Recognition and Language Identification for Ethiopian Languages
March 2026
We present Ethio-ASR, a suite of multilingual CTC-based automatic speech recognition (ASR) models jointly trained on five Ethiopian languages: Amharic, Tigrinya, Oromo, Sidaama, and Wolaytta. These languages belong to the Semitic, Cushitic, and Omotic branches of the Afroasiatic family, and remain severely underrepresented in speech technology despite being spoken by the vast majority of Ethiopia’s population. We train our models on the recently released WAXAL corpus using several pre-trained speech encoders and evaluate against strong multilingual baselines, including OmniASR. Our best mod... -
AmharicStoryQA: A Multicultural Story Question Answering Benchmark in Amharic
February 2026
With the growing emphasis on multilingual and cultural evaluation benchmarks for large language models, language and culture are often treated as synonymous, and performance is commonly used as a proxy for a models understanding of a given language. In this work, we argue that such evaluations overlook meaningful cultural variation that exists within a single language. We address this gap by focusing on narratives from different regions of Ethiopia and demonstrate that, despite shared linguistic characteristics, region-specific and domain-specific content substantially influences language e... -
Afri-MCQA: Multimodal Cultural Question Answering for African Languages
January 2026
Africa is home to over one-third of the world’s languages, yet remains underrepresented in AI research. We introduce Afri-MCQA, the first Multilingual Cultural Question-Answering benchmark covering 7.5k Q&A pairs across 15 African languages from 12 countries. The benchmark offers parallel English-African language Q&A pairs across text and speech modalities and was entirely created by native speakers. Benchmarking large language models (LLMs) on Afri-MCQA shows that open-weight models perform poorly across evaluated cultures, with near-zero accuracy on open-ended VQA when queried in ...
Projects
Fetching repositories from GitHub...