Research Engineer

Research Engineer

Team: Research

Location: Bhaktapur

Role Overview

As a Research Engineer at SAGEA, you will work at the boundary of research and engineering to design, implement, and evaluate reasoning-driven language models. This role focuses on turning research ideas into working systems through rigorous experimentation, efficient implementation, and careful evaluation.

You will collaborate closely with research scientists and engineers to prototype new model architectures, run controlled experiments, and improve model efficiency and deployment readiness.

What You Will Do

• Design and implement model architectures and training pipelines for language and reasoning systems
• Run experiments to evaluate model behavior, efficiency, and scalability
• Prototype and iterate on research ideas, turning concepts into measurable results
• Improve inference efficiency and model performance under constrained environments
• Analyze experimental results and contribute to internal research reports and documentation
• Support reproducibility by maintaining clean experiment pipelines and versioned results
• Collaborate with engineering teams to transition research outputs toward deployment

What We Are Looking For

• Strong foundation in machine learning and deep learning
• Solid programming skills, primarily in Python
• Experience working with neural networks and modern ML frameworks
• Ability to reason about experiments, metrics, and tradeoffs
• Comfort working in fast moving research environments with incomplete information

Nice to Have

• Experience with language models, transformers, or reasoning architectures
• Familiarity with training large models or optimizing inference
• Exposure to research workflows, ablation studies, and benchmarking
• Experience reading and implementing research papers

Why This Role Matters at SAGEA

At SAGEA, Research Engineers are central to how ideas become systems. This role ensures that research is not isolated from engineering and that experimental advances are grounded in implementation, efficiency, and real-world constraints.