Duressa S.

Key Skills

Software

Don't disclose

Other

Top Subject Matter

Machine learning

AI-driven recommendations

scalable backend infrastructure for ML training workflows

Top Data Types

Text

Video

Top Task Types

Fine-tuning

Text Generation

Freelancer Overview

Software Engineer (ML & Backend) at Future Labs (Adama, Ethiopia). Brings 4+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Don't disclose and Other. Education includes Bachelor of Science, Adama Science and Technology University and Advanced Software Engineering, A2SV Academy. AI-training focus includes data types such as Computer Code, Programming, and Text and labeling workflows including Fine-tuning, Translation, and Localization.

Labeling Experience

Project: engineX-transformerBT (Jan 2025 - Present)

OtherText

Created a custom Transformer model for English-to-Afaan Oromo translation, involving dataset preparation and training-related pipeline engineering. Designed tokenization and learning-rate scheduling components aligned with foundational Transformer training approaches. Contributed to NLP improvements for underrepresented African languages through model experimentation and evaluation. • Trained a Transformer from scratch in PyTorch for translation. • Implemented tokenization pipelines and a learning rate scheduler. • Built components supporting translation/localization tasks. • Worked on NLP advancement efforts for low-resource language data.

2025 - Present

Software Engineer (ML & Backend) at Future Labs (Adama, Ethiopia)

Don't discloseFine-tuning

Worked on ML-driven content recommendation systems and model development practices that support training and iterative improvement of AI pipelines. Implemented microservices and real-time features that provide infrastructure for collecting, processing, and serving AI/behavioral signals used in model training workflows. Optimized data retrieval and system performance to enable efficient experimentation and downstream ML training iterations. • Built an AI-driven content recommendation backend with scalable ML support. • Designed microservices for independent scaling of user, posts, video, search, and messaging services. • Reduced API response times through a multi-layer caching strategy for faster data access. • Improved real-time messaging latency using WebSockets for near-real-time data flow.

2024 - Present

Project: gamma-decoderGPT (Dec 2024 - Feb 2025)

OtherTextText Generation

Built a decoder-only Transformer (GPT-like) model for autoregressive text generation, requiring training pipeline development. Developed preprocessing with subword tokenization and implemented custom learning-rate scheduling to support training runs. Provided an interactive Flask web interface for generated text, enabling iterative testing and evaluation of the model. • Trained a GPT-like decoder-only Transformer from scratch in PyTorch. • Implemented a training pipeline using subword tokenization. • Added positional embeddings and multi-head attention. • Deployed a Flask interface for interactive generation and review.

2024 - 2025

Education

Work History