Welcome
Ian Kolaja is a machine learning engineer and researcher working at the intersection of ML systems, scientific computing, and AI-integrated tooling. He earned his PhD in Nuclear Engineering from UC Berkeley, with a designated emphasis in Computational and Data Science and Engineering, and also completed his undergraduate degree at Berkeley after transferring from Fullerton College. Ian is a strong advocate for transfer students and for those impacted by the foster youth system.
Academic Research
Ian's dissertation addressed key challenges in monitoring, modeling, and operating advanced nuclear reactor systems. He adapted bent crystal diffraction (BCD) spectrometers as energy-selective gamma-ray filters for rapid fuel characterization, coupling them with ML models to infer pebble properties such as burnup and neutron fluence. He also developed LSTM-based sequence models to forecast reactor state over time, enabling optimization of startup and running-in sequences. Extensions to molten salt reactors demonstrated real-time tracking of plutonium production for safeguards applications.
Separately, Ian built NuGrade, a data quality analysis platform for nuclear cross section data that combines statistical evaluation pipelines, LLM-based agents, and a RAG system for semantic retrieval of experimental reports, applying modern NLP and AI tooling to longstanding challenges in nuclear data curation.
Ian's dissertation will be available on ProQuest on September 30th, 2026.
Industry Experience
Before his PhD, Ian worked at Kairos Power, where he was the first developer of the Kairos Power Advanced Core Simulator (KPACS). KPACS supported equilibrium and transient core studies for the company's generic fluoride-salt-cooled high-temperature reactor (gFHR) benchmark and contributed to early design analyses for KP-Hermes 1.
Outside of Nuclear
Ian has a wide range of creative pursuits including music production and filmmaking. Before pursuing science in community college, he wanted to be a film director. He participated and won awards at a handful of student film festivals in Southern California. He also recently has released his first EP, a collection of indie pop songs.
Projects
NuGrade
Ian designed and built NuGrade, a nuclear data quality analysis web tool for assessing cross section data from EXFOR and evaluations such as ENDF. Data is graded on a reaction-channel basis using a metric that combines agreement between experimental data and evaluations with energy coverage. Users can access an interactive chart of nuclides, adjust scoring parameters, and browse data for specific nuclides and reactions.
NuGrade features an LLM agent capable of analyzing the data and computed metrics. A natural language processing pipeline allows experimental reports from EXFOR to be incorporated into this analysis: SciBERT is used to compute sentence-level embeddings, and a retrieval-augmented generation (RAG) framework enables the agent to find, reference, and discuss text from relevant reports. Missing measurement uncertainties can be imputed using K-nearest neighbors by identifying the most similar experiments that have reported uncertainty values.
The NuGrade repository is available on GitHub, and it may be publicly hosted in the future with an accompanying report or article.
Pebble Bed Reactor State Forecasting
Ian addressed challenges in PBR operation using a time series analysis approach. Reactivity control in PBRs is difficult due to the coupled, nonlinear relationships between different control mechanisms, which also act on different time scales. In-core measurement is limited in high-temperature environments, making it essential to leverage operation history and discharge fuel measurements to infer the state of the reactor core.
Reactor State Forecasting Model Architecture
Ian designed an ML system for forecasting reactor behavior using Long Short-Term Memory (LSTM) neural networks trained on simulation data generated with Monte Carlo methods. Synthetic measurements intended to mirror what operators could feasibly collect in practice were constructed and, coupled with operation history, served as the noisy input to the model, which achieved an R² of 0.991 on held-out data.
The LSTM model was coupled with a simulator to predict the impact of candidate control actions, allowing the model to steer the simulator from low to high power while maintaining criticality. This approach offers reactor operators a fast way to assess the impact of different control strategies before committing to an expensive full simulation.
Planned System Design for Application of Models in Real Reactor
An ArXiv manuscript describing this work is available.
Bent Crystal Diffraction Spectrometry
Ian investigated the challenge of rapidly measuring fuel pebbles discharged from PBRs. These pebbles are highly radioactive and must be processed quickly to avoid imposing additional operating constraints on the reactor. One approach for enhancing this process is the use of bent crystal diffraction (BCD) spectrometers, which act as energy-selective photon filters. By filtering specific gamma rays or narrow energy bands, virtually all Compton scattering and dead time can be eliminated, enabling much more accurate inference of pebble properties including burnup, number of passes, actinide concentrations, residence time, and neutron fluence.
This work resulted in a provisional patent and a peer-reviewed publication in Annals of Nuclear Energy, which is available online.
Schematic of BCD Spectrometer
Publications
"Burnup measurement using bent crystal diffraction spectrometers for pebble
bed reactor"
I. Kolaja et al., Annals of Nuclear Energy, vol. 233, Aug. 2026, url: https://doi.org/10.1016/j.anucene.2026.112263
"Machine Learning Prediction of Pebble History in Pebble Bed
Reactors"
I. Kolaja et al., in
Proceedings of the International Conference on Physics of Reactors (PHYSOR 2024), Apr. 2024, pp.
1437-1446.
doi: 10.13182/PHYSOR24-43890.
"Neutronics, thermal-hydraulics, and multi-physics benchmark models for a
generic pebble-bed fluoride-salt-cooled high temperature reactor (FHR)"
N. Satvat et al., Nuclear Engineering and Design, vol. 384, no. C, Nov. 2021
doi: 10.1016/j.nucengdes.2021.111461.
"First Direct Measurements of Superheavy-Element Mass Numbers,"
J. M. Gates et al., Phys. Rev. Lett., vol. 121, no. 22, p. 222501, Nov. 2018,
doi: 10.1103/PhysRevLett.121.222501.
Preprints
"Predicting and forecasting reactivity and flux using long short-term
memory
models in pebble bed reactors during run-in"
I. Kolaja et al., arXiv, Nov. 2025, url: https://arxiv.org/abs/2511.05118
Curriculum Vitae
You can download a PDF copy of my CV here.
Presentations
UC Berkeley Colloquium: PhD Exit Talk
Advanced Fuel Measurement and Operation Modeling for Pebble Bed Reactors
I. Kolaja, UC Berkeley, November 2025
Slides
Recording
PhD Qualifying Examination
I. Kolaja, UC Berkeley, May 2024
Slides
Machine Learning Prediction of Pebble History and Nuclide Concentration in
PBRs
I. Kolaja et al., International Conference on Physics of Reactors (PHYSOR 2024), Apr. 2024
Session: Machine Learning and Artificial Intelligence for Reactor Physics: I
Slides
Resources
NE150/215M: Introduction to Nuclear Reactor Theory
UC Berkeley Nuclear Engineering Department, Spring 2022
GSI Discussion Slides by Ian Kolaja
Note: I joined the course a few weeks in, so weeks 1-3 are missing. These slides have not been edited since
the course, so if you see any mistakes please let me know!
Discussion 4: Criticality
Discussion 5: Neutron
Transport
Midterm 1 Review:
Slides
Midterm 1 Review:
Worksheet
Midterm 1 Review:
Solutions
Discussion 6: Diffusion Equation
Pt. 1
Discussion 7: Diffusion Equation
Pt. 2
Discussion 8: Diffusion Equation
Pt. 3
Discussion 9: Multigroup
Diffusion
Discussion 10: Reactor Kinetics
and Long-Term Core Behavior
Midterm 2 Review:
Slides
Midterm 2 Review:
Worksheet
Midterm 2 Review:
Solutions
Resources for Graduate Students
Curated for Nuclear Engineering students
Best Practices for Scientific Computing (by Greg
Wilson et al.)
Media
[Video] PhD Exit Talk at UC Berkeley Nuclear Engineering Colloquium: Advanced
Fuel Measurement and Operation Modeling for Pebble Bed Reactors
By Ian Kolaja
Youtube, November 2025
[Video]
ARCO: Making Advanced Reactors Smarter
By Ian Kolaja, Laura Shi, and Eddie Bird
Vimeo, June 2019
[Article] How UC Berkeley researchers are making nuclear
reactors smarter
By Laura Shi, Ian Kolaja, and Eddie Bird
Fung Institute for Engineering Leadership, August 2019
[Video]
Nuclear Values
By Ian Kolaja
Vimeo, March 2018