AI Research Engineer

Habtamu Asefa

I work on Amharic speech models and build AI tools for learning, research, and writing.

CTO of PrepX, an education startup. Currently building text-to-speech systems at Liqawnt. Based in Addis Ababa.

Habtamu Asefa
Addis Ababa, Ethiopia

Selected work

Project details

AI research workspace · In development

Socratic Loop

A desktop workspace for reading research papers, taking notes, and drafting writing with an AI collaborator. The agent finds related work, compares papers, and proposes edits grounded in sources from the workspace.

TypeScript · Electron · React · Tiptap · PDF.js

Project details

Android AI keyboard · In development

Poetic Keyboard

An Android keyboard that fixes grammar, rewrites replies, and polishes tone where you type. Supports translation, draft continuation, and custom AI actions with streamed results.

Kotlin · Android IME · Coroutines · Server-sent events

Project details

Speech research · 2026–present

Liqawnt

Amharic text-to-speech work: data preparation, text normalization, VITS fine-tuning, synthesis, and evaluation. Speech recognition and speech-to-speech are longer-term research directions.

Python · PyTorch · torchaudio · Transformers

Research details

Open source · 2025–present

react-native-ajora

A React Native library for AI chat, with AG-UI integration, streamed messages, tool-call rendering, conversation state, and client-side tools. Used in PrepX.

TypeScript · React Native · AG-UI

Source code

Learning project · 2026

myTorch

A small Python implementation of scalar automatic differentiation, MLP layers, SGD, and a byte-level BPE tokenizer.

Python · Autograd · Tokenization

Source code

Biomedical engineering thesis · 2024

UlcerGuard

A pressure-sensing footwear prototype for identifying high-pressure regions associated with diabetic foot-ulcer risk, with a companion mobile application.

Sensors · React Native

Mobile source code

Experience

Liqawnt

AI research and engineering

2026–present

Developing an Amharic TTS pipeline and fine-tuning workflow. Research interests: low-resource speech, multimodal models, and efficient inference.

PrepX

CTO

2024–present

Lead product and engineering for the study app. Publicly launched in March 2026; maintain the mobile app, backend services, and AI integrations.

Freelance and agency work

Software Engineer · US and EU clients

2023–present

Delivered React Native features for Dymnd, including timezone-sensitive habit tracking and authentication fixes. Maintained backend services and deployment infrastructure for a US bar-exam preparation platform.

Temaribet

Founder

2019–2025

Founded and operated a tutoring marketplace. Managed tutor onboarding and family matching; selected into Reach for Change and Orbit Innovation Hub.

Technical skills

Languages
Python, TypeScript, JavaScript, SQL
Machine learning
PyTorch, torchaudio, Hugging Face Transformers
Web and mobile
React, Next.js, React Native, Expo
Backend and data
FastAPI, NestJS, PostgreSQL, SQLite, SQLAlchemy, Prisma
Search and AI tools
Qdrant, Typesense, FastEmbed, Mastra, AG-UI
Infrastructure
Linux, Docker, Nginx, AWS EC2, S3, GitHub Actions

Education

Addis Ababa University / AAiT

BSc in Biomedical Engineering · 2024

Thesis: UlcerGuard, pressure-sensing footwear for diabetic foot-ulcer prevention.

Contact

For speech research, AI engineering, or software collaborations.