2026
AI-research-feedback
A collection of Claude Code skills for academic research review. These tools were developed by Claes Bäckman (economics). I modified them for cog/neuro science. It is useful to compare regenerated results from a replication package against a manuscript draft, or to simulate feedbacks.
March Madness 2026 Forecasting Engine
Built an end-to-end public-facing prediction and diagnostics engine for the 2026 NCAA tournament. Fetched BartTorvik, WarrenNolan and NCAA historic data, tested predicting models, created brackets portfolio, and diagnosed predictions after each round. It was built in 3.5 days using an earlier version of SUPLEX.
SUPLEX-agentic-workflow
Developed a control layer for bounded LLM-assisted repository work, built around explicit handoffs, execution reports, scoped task boundaries, and continuity artifacts. It is used to structure changes so delegation of tasks to LLMs does not erase provenance, scope discipline, or reviewability. Now used in each of my most complex projects.
Ten simple rules for achieving computational reproducibility in neuroscience
Gaurav Mahajan and I are formalizing the infrastructures we built for the UMD SLD Lab into 10 simple rules for achieving computational reproducibility in neuroscience. We emphasize simple, low-cost decisions that can be implemented early and scaled as projects become more complex. The manuscript comes with a companion document specifying codelines to implement each rule.
2025
Neurocomputational Mechanisms of Social Learning
Social learning is crucial for acquiring knowledge about the world, deciding how to act in social situations and learning consequences of actions without direct experience. Yet social learning vary widely across individuals, including in disorders associated with deficits in social function, such as autism and social anxiety. We need a robust characterization at the individual level. In a large-scale fMRI study, we seek to better characterize this heterogeneity in processes and brain correlates. We use a social learning battery that crosses sociality and importance of learning across tasks of risk/ambiguity, trust learning and observational learning.
Neurodesign-plus
Developed an open-source python package to optimize fMRI design in typical complex, decision-making experimental setups. I led the project and Atharv Umap did the heavy coding work. The package Addresses shortcomings in existing solutions for optimizing trial or time allocation during design; allows specifying intricate, random or path-dependent designs. It is built on top of Poldrack lab's neurodesign.
fMRI Fitlins Templates
Built reusable templates for fMRI interactive and slurm work using Python-based fitlins pipelines. Packages analysis, logging and plotting using nilearn and fitlins wrappers.
fMRIPrep QC guide and logging
Designed a guide and a logging system for assessing quality of data preprocessed using fMRIPrep.
2024
Individual Differences In Dynamic Belief Updating During Trust Learning
Trust is defined as the willingness to be vulnerable to another being on the basis ofpositive expectations of their intentions and behaviors. In repeated interactions, learning to trust others involves cognitive processes that integrate uncertainty, context, and potential betrayal. Different strategies can guide trust behavior, such as heuristics, learning, or beliefs updating. Individuals differ in how they deploy cognitive processes, which may impact their strategies. In a dynamic trust learning task that can be solved using a variety of strategies, we characterize systematically this heterogeneity, identifying different profiles of participants mobilizing different cognitive processes.
Personal Website
The site you are reading: a Jekyll (built from al-folio) research-facing website.
Polarization and social learning in social network
I initialized a study on the formation of beliefs in social networks with Jean-Claude Dreher, Frederic Moisan and Alain Barrat. Gaël Carniel is now conducting the study. At the collective level, the behavior of each individual contributes to the evolution of the architecture of social networks in which we are embedded. In turn, those networks affect the beliefs we form about the state of the world. With this study, our main objective is to investigate how information propagation impacts the formation of consensus/dissensus/polarization, reflected by the formation of communities, and the role of social learning in large groups.
2023
Cookiecutter Neuro Research Project
Created a cookiecutter template for psyc/neuro research. It's a reusable scaffold for starting new research projects with a clearer architecture from the outset. The template is oriented toward computational neuroscience work and is designed to make setup, organization, and handoff of analysis projects across Python, R, and MATLAB-based workflows. Highly recommended to initialize any research project.
2020
ReceivingNews - Modeling Information-seeking Post News Evaluation
Project ReceivingNews (Metacognition biases information seeking in assessing ambiguous news, published in Communications Psychology) in which we show that humans have limited metacognitive ability when facing ambiguous news; that this metacognition drives information-seeking; and that news ambiguity (news content imprecision and propensity to polarize opinions) explain against any other variable individuals' likelihood of misjudgment. Our results underscore the importance of metacognitive abilities in mediating the relationship between assessing ambiguous information and the decision to seek or avoid more information.
SendingNews - Neurocomputational Mechanisms of Infering Others' Preferences For Information
Project SendingNews (Neurocomputational mechanisms of sharing debunking information about ambiguous news) in which we show that humans infer others' preferences for information with a Bayesian updating rule that integrates first-order beliefs about news accuracy with second-order beliefs about recipients’ preferences. Second-order beliefs specifically include inferences from both social distance and learning about Receivers’ general reception tendencies. Neuroimaging results reveal that the ventromedial prefrontal cortex tracked first-order beliefs while the temporoparietal junction and dorsomedial prefrontal cortex tracked second-order beliefs.
2019
fMRI SPM Templates
Built reusable templates for fMRI work: SPM-based pipelines for analysis structure, preprocessing logic, and project setup.
2018
Sex-Money - Testosterone Causes OFC-Amygdala Decoupling
We hypothesized that testosterone administration may: 1) increase posterior lateral orbitofrontal cortex activity, previously observed to be engaged more with erotic as compared to monetary rewards in healthy young men; (2) decrease the functional coupling between the medial part of the orbitofrontal cortex and the amygdala while anticipating rewards. I showed that testosterone specifically increased incentive behavior related to erotic stimuli as compared to monetary rewards; that this behavioral interaction effect was associated with a higher association between the relative motivational value for erotic as compared to monetary cues in the ventral striatum; and that testosterone injection reduced the functional coupling between the ventromedial prefrontal cortex and the amygdala while anticipating both primary and secondary rewards, showing testosterone affects limbic-prefrontal connectivity during reward processing.