Aavash Dahal

I am a Master's student in Data Science and Artificial Intelligence at Saarland University, Germany, supported by the Zuse School ELIZA scholarship. Alongside my studies, I am a Research Assistant (HiWi) at the Fraunhofer Institute. I hold a Bachelor's degree in Computational Mathematics from Kathmandu University, Nepal.

My professional journey includes working as a Software Developer at Ridgehead Inc. and a Machine Learning Research Internship at Wiseyak, where I worked on large language models and modern deep learning pipelines.

profile photo
About me

My research focuses on making large language models and LLM-based agents more trustworthy. I am particularly interested in factuality, uncertainty estimation, and the reliability of agentic systems: when can we trust what a model says or does, and how can a system recognize when it should not be trusted?

Broadly, I am drawn to problems at the intersection of:

My long-term goal is to build AI systems whose outputs can be verified and whose confidence can be relied upon, grounded in sound statistical and mathematical foundations.

Honors & Awards 🏆
ELIZA Scholarship Recipient 2025 - Present
Merit-based scholarship awarded to selected Master's students in Data Science and Artificial Intelligence at Saarland University, recognizing academic performance and research potential.
Current Focus
Experience & Projects
Research Assistant (HiWi), Fraunhofer Institute
Oct 2025 – Present

Contributing to research projects involving agentic AI systems and large language models, including experimentation, evaluation, and pipeline development.

Software Developer, Ridgehead Inc.
2023 – 2025

Engineered scalable production systems and data-driven web applications, strengthening my engineering foundation for building robust ML systems.

Machine Learning Research Intern, Wiseyak
2023 (3 months)

Developed and optimized deep learning models and pipelines for LLM-based applications, focusing on performance, reliability, and evaluation.

Writing
Learning to Learn by Gradient Descent by Gradient Descent 2026
A from-scratch PyTorch implementation of a meta-learned LSTM optimizer, stress-tested against SGD and Adam across three scenarios.
Events & Training
Berlin Summer School on Artificial Intelligence and Society (BIFOLD) Sep 2026
Participant. Four-day summer school at TU Berlin on the responsible design of AI-driven autonomous systems, covering trustworthy machine learning, robustness, explainability, and governance.
Community & Activities