I'm Amanuel Gizachew, an AI Research Engineer 🧠 from Ethiopia 🇪🇹. 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.

• 🔭 Current Focus: Developing AfriNLLB, a series of lightweight translation models for 15+ African languages.
• 🌱 Learning: Currently diving into Cross-Platform Development with Flutter and exploring advanced Reasoning Models (HRM).
• 🤝 Collaboration: I’m eager to partner on NLP for low-resource languages, Computer Vision (CV), and initiatives for AI social good in Africa.
• 🏆 Achievements: * 1st Place: Addis Ababa Innovation Competition (AI for Medical Systems).
  • 2nd Place: Akaki Kality Innovation Competition (Trilingual Native Chatbot).
• 🤔 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.