Bedru Yimam Ahmed

Department of Information Technology

KIoT, Wollo University

Kombolcha, Ethiopia

I am a Lecturer and Researcher in the Department of Information Technology at Wollo University (Kombolcha Institute of Technology - KIoT), Ethiopia. I hold a Master of Science (MSc) in Information Technology from Bahir Dar University.

My research investigates the computational frontiers of Low-Resource Natural Language Processing and Speech Technology, with a special focus on Semitic languages and Ge’ez script computing (Amharic, Tigrinya). I focus on making modern AI systems equitable, compute-efficient, and practically deployable in under-represented language communities.

Core Research Pillars

LLM Tokenization & Efficiency

Quantifying and resolving the 9.6x cross-lingual Token Tax, subword fragmentation, and byte-fallback collapse in English-centric LLMs processing Ethiopic script.

Parameter-Efficient Fine-Tuning (PEFT)

Low-rank adaptation (LoRA) of compact open-weight models (Qwen, Gemma, Aya) on modest hardware, audited via Bits-per-Character (BPC) and paired bootstrap confidence intervals.

Clinical & Domain-Specific RAG

Designing offline-capable Retrieval-Augmented Generation architectures with Hybrid BM25 + Dense FAISS and Reciprocal Rank Fusion (RRF) for maternal healthcare (ANC guidelines).

Dialect Speech Recognition (ASR)

Adapting foundation speech models (OpenAI Whisper) to regional Ethiopian dialects—specifically Wollo Amharic phonology and acoustic variants.

Academic Collaborations & Prospective Opportunities

I actively welcome collaborations with research labs, industry researchers, and academic institutions working on low-resource language technologies, speech recognition, and African NLP.

I am also actively exploring PhD positions and research fellowships focused on Multilingual Foundation Models, Parameter-Efficient Adaptation, and Speech Intelligence.

Feel free to connect via Email, Google Scholar, or LinkedIn.

news

Oct 10, 2026	📊 Released the Amharic Tokenization Benchmark, evaluating subword fertility and the 9.6x Token Tax across 9 LLMs on FLORES-200. Explore the code and benchmark.
Jul 28, 2025	🎉 Our paper “Automatic Speech Recognition for African Low-Resource Languages: Challenges and Future Directions” was presented at ACL AfricaNLP 2025 in Vienna, Austria! Check out the ACL Anthology Paper.

selected publications

AfricaNLP 2025

Automatic Speech Recognition for African Low-Resource Languages: Challenges and Future Directions

Sukairaj Hafiz Imam, Babangida Sani, Dawit Ketema Gete, and 6 more authors

In Proceedings of the Sixth Workshop on African Natural Language Processing (AfricaNLP 2025), Jul 2025

Abs HTML PDF

Automatic Speech Recognition (ASR) technologies have transformed human-computer interaction; however, low-resource languages in Africa remain significantly underrepresented in both research and practical applications. This study investigates the major challenges hindering the development of ASR systems for these languages, including data scarcity, linguistic complexity, limited computational resources, acoustic variability, and ethical concerns surrounding bias and privacy. The findings highlight the importance of interdisciplinary collaboration and sustained investment to advance low-resource speech technology.
Sci Rep

Multimodal based Amharic fake news detection using CNN and attention-based BiLSTM

Alemu Desu Geto, Eshete Derb Emiru, Nuru Endris Seid, and 2 more authors

Scientific Reports, Oct 2025

HTML PDF
ResearchGate

Whispering in Amharic: Fine-tuning Whisper for Low-resource Language

Dawit Ketema Gete, Bedru Yimam Ahmed, Tadesse Destaw Belay, and 1 more author

Preprint, Oct 2024

HTML