Nawaf Alampara
Doctoral Researcher
Friedrich-Schiller-Universität Jena
I am second-year PhD student, working with Dr. Kevin Maik Jablonka. I'm building machine learning systems to speed up scientific research, and I love projects that involve both research and building the tooling that enables and accelerates that research. Lately, I’ve been analyzing general-purpose AI models/systems to understand their limitations in scientific applications—where they fail—and interpreting them to uncover why they fail. My goal is to use these insights to design AI systems that aren’t just impressive on benchmarks but truly impactful for advancing science and research.
News
February 2026
Joined Lila Sciences
I joined Lila Sciences as an AI resident. I would be working from Boston 🎉
May 2025
Accepted for Google Summer of Code 2025
I will contribute to DeepMind. Effort will be towards evaluating scientific reasoning capabilities of Gemini models. Read more about my project.
Publications
arXiv 2026
AI scientists produce results without reasoning scientifically
Nawaf Alampara, Martiño Ríos-García, .., N. M. Anoop Krishnan, Kevin Maik Jablonka
25,000 AI scientist experiments reveal that LLMs ignore gathered evidence, never update beliefs, and harness has less impact compared to the underlying model.
Chemical Reviews 2026
General-Purpose Models for the Chemical Sciences: LLMs and Beyond
Nawaf Alampara, Anagha Aneesh, Martiño Ríos-García, Adrian Mirza, Mara Schilling-Wilhelmi, ..., Kevin Maik Jablonka
Theory, concepts, application and thought behind "general purpose models" for chemical sciences.
Nature Computational Science 2025
⭐ Neurips AI4Mat
Probing the limitations of multimodal language models for chemistry and materials research
Nawaf Alampara, ..,Kevin Maik Jablonka
Multimodal benchmark for chemistry/materials science for AI with ablations to interpret the limitations
Nature Chemistry 2025
A framework for evaluating the chemical knowledge and reasoning abilities of large language models against the expertise of chemists
Adrian Mirza, Nawaf Alampara, ..,Kevin Maik Jablonka
First comprehensive benchmark for chemistry-specific AI capabilities, evaluating chemical knowledge, intuition, and reasoning of LLMs against human chemists.
AI4Mat-Vienna 2024 2024
⭐ spotlight (oral)
MatText: Do Language Models Need More than Text & Scale for Materials Modeling?
Nawaf Alampara, Santiago Miret, Kevin Maik Jablonka
Revealing Transformer models' (IFT and trained from scratch) limitations in capturing 3D geometric information crucial for materials modeling.
Experience
AI Resident (Intern) — Lila Sciences
AI agents | Test time strategies | Long context scientific agent environments
PhD Researcher — Friedrich-Schiller-Universität Jena
Advisor: Dr. Kevin Maik Jablonka
AI Research Contractor (Part-time) — Stability AI
Dataset curation | Benchmarking
Principal Engineer — QpiVolta Technologies
Material simulation using geometric deep learning models | Software development
Research Engineer — QpiAI Technologies
Real-time video analytics | Computer vision
Education
Friedrich-Schiller-Universität Jena, Germany
PhD Machine Learning for Science
Advisor: Dr. Kevin Maik Jablonka
Indian Institute of Technology Bombay, India
MSc Energy Science
Advisor: Prof. K R Balasubramaniam
Thesis: Defects and Dopants in Cu₂O - DFT study