Hewan Shrestha

Doctoral Researcher
GESIS – Leibniz Institute for the Social Sciences

Portrait of Hewan Shrestha

I am a Doctoral Researcher at GESIS – Leibniz Institute for the Social Sciences in Cologne, Germany. My work is about how we judge whether an AI system is actually any good. I build the datasets, evaluation methods and documentation that make claims about language models and AI research assistants possible to check. This work is part of NFDI4DS, Germany’s national research data infrastructure for data science and artificial intelligence.

Before this I spent three years as a Research Assistant in the Interdisciplinary Institute for Societal Computing (I2SC) led by Prof. Dr. Ingmar Weber, estimating mobility in conflict-affected regions from low-resolution satellite imagery. I hold an M.Sc. in Visual Computing from Saarland University, where my thesis on multimodal misinformation detection was supervised by Prof. Dr. Dietrich Klakow and Dr. Usman Naseem. I received my Bachelors degree in Computer Science and Engineering from Jawaharlal Nehru Technological University Anantapur, India, and completed my Bachelors thesis in collaboration with Innopolis University, Russia, supervised by Prof. Dr. Manuel Mazzara.

  1. GESIS
    Doctoral Researcher
    2026 – Present

  2. I2SC
    Research Assistant
    2023 – 2026

  3. Saarland University
    M.Sc. Visual Computing
    2023 – 2026

  4. Innopolis University
    Research Intern
    2022

  5. JNTUA
    B.Tech.
    2018 – 2022

Research Interests

Evaluation & benchmarking practices for AI

Documenting what benchmark datasets actually cover and where they are biased, and testing models for robustness and reproducibility rather than headline accuracy.

Evaluating research co-pilots

Defining the tasks AI research assistants are asked to perform, and the protocols, metrics and user studies needed to judge how well they do them.

Domain-specific models for scholarly search

Adapting language and embedding models to social science language, and measuring whether that genuinely improves search over scientific literature and research data.

Multimodal & low-resource NLP

Building text-and-image models for languages with little training data, and the benchmarks to measure them, so tools such as automated fact-checking work beyond English.

News

Nov 2026

Started a Research Associate/Doctoral Researcher position at GESIS – Leibniz Institute for the Social Sciences in Cologne, Germany.

Oct 2026

Concluded my time as a Research Assistant at the Interdisciplinary Institute for Societal Computing (I2SC), Saarland University.

Sep 2026

Completed my M.Sc. in Visual Computing at Saarland University, Germany.

Apr 2025

Our workshop paper on a Weak Supervision Learning Approach Towards an Equitable Mobility Estimation was accepted at ICWSM 2025.

Dec 2023

Our paper about LLM-Based Recommender Systems was accepted at the 33rd Annual Workshop on Information Technologies and Systems (WITS'23).

May 2023

Started working as a Research Assistant in the Societal Computing research group at Saarland University, Germany.

Apr 2023

Started Masters in Visual Computing at Saarland University, Germany.

Feb 2023

Our paper about Melanoma Segmentation was accepted to Springer Multimedia Tools and Applications.

Oct 2022

Our paper about Face Mask Recognition was accepted to the Springer Lecture Notes in Networks and Systems book series.

Jun 2022

Graduated with a B.Tech. in Computer Science & Engineering from Jawaharlal Nehru Technological University Anantapur, India.

Dec 2021

Invited to be a reviewer for The Journal of Supercomputing.

Jun 2021

Our paper about Brain Tumor Detection was accepted to the Springer Advances in Intelligent Systems and Computing book series.

Jul 2020

Our paper about sentiment analysis was accepted to the International Journal of Speech Technology.

Publications

A Weak Supervision Learning Approach Towards an Equitable Mobility Estimation

Theophilus Aidoo, Till Koebe, Akansh Maurya, Hewan Shrestha, Ingmar Weber

ICWSM Workshop 2025

  Paper   Code

LLMRS: Unlocking Potentials of LLM-Based Recommender Systems for Software Purchase

Angela John, Theophilus Aidoo, Hamayoon Behmanush, Irem B. Gunduz, Hewan Shrestha, Maxx Richard Rahman, Wolfgang Maaß

WITS 2023

  Paper   Code   Slides

For the full list of publications, see my Google Scholar profile.