I'm Amanuel Gizachew, an AI Research Engineer ๐ง from Ethiopia ๐ช๐น. My mission is to pioneer advancements for Amharic and other low-resource African languages using state-of-the-art Machine Learning.
I'm currently building a startup focused on African LLMs in collaboration with the Ethiopian Artificial Intelligence Institute. I believe in AI for social good and am dedicated to solving real-world challenges through intelligent, localized solutions.
- ๐ญ Current Focus: Developing AfriNLLB, a series of lightweight translation models for 15+ African languages.
- ๐ฑ Learning: Currently diving into Cross-Platform Development with Flutter and exploring advanced Reasoning Models (HRM).
- ๐ค Collaboration: Iโm eager to partner on NLP for low-resource languages, Computer Vision (CV), and initiatives for AI social good in Africa.
- ๐ Achievements: * 1st Place: Addis Ababa Innovation Competition (AI for Medical Systems).
- 2nd Place: Akaki Kality Innovation Competition (Trilingual Native Chatbot).
- ๐ค Challenge: Actively seeking ways to optimize compute power for training large models in resource-constrained environments.
- AfriNLLB (2026): Efficient Translation Models for African Languages. (AfricaNLP 2026)
- Vision-Language Alignment (2026): Steering Vision Models towards Subjective Concepts with Language. (CHI '26 Workshop)
AI/ML & Data Science:
Programming & Frameworks:
- Amharic Qwen Implementation: Rebuilt the Qwen architecture from scratch using PyTorch specifically for Amharic text generation.
- Reflection Tuning: Developed a novel finetuning method for low-resource languages, yielding a 10% performance improvement on Llama models.
- Amharic Vision-Language Model: Implemented "See and Tell" research for native language image captioning.
- HRM Text Generator: A 27M parameter model based on Hierarchical Reasoning architectures.
