Open to opportunities

Sara Kamali

@sarakamali

Message

Machine-learning researcher designing deep sequence models for robust time-series representations.

Spain

Message

What I'm looking for

I’m looking to build and validate robust, interpretable deep learning models for noisy, non-stationary time-series—especially biosignals—using rigorous experimental design, explicit robustness checks, and reproducible open-source code.

I am a machine-learning researcher with a Ph.D. (Cum Laude) in Computer Science and a peer-reviewed publication record across journals, conference proceedings, preprints, and open-source scientific software. I design, implement, and evaluate deep-learning models for noisy, non-stationary, small-sample time-series data, carrying work from problem framing through experimentation, analysis, and reproducible code release.

My hands-on research focuses on deep sequence and attention-based architectures, including a self-attentive BiLSTM for temporal decoding and transformer-based time-series models such as PatchTST, trained on GPU HPC clusters. I care as much about whether a result generalizes as about the model that produced it, so I make rigorous validation standard—nested and leave-one-subject-out cross-validation, permutation testing, stability selection, surrogate analysis, and explicit robustness checks.

Across roles at the Autonomous University of Madrid and in visiting research at UC San Diego, I’ve applied these principles to biomedical and physiological time-series (EEG/EMG), including clinical dementia EEG datasets and stimulation- and response-characterization workflows. I bring strong mathematical foundations in linear algebra, probability, and statistics, and I enjoy building end-to-end pipelines that produce interpretable, physiologically plausible insights while keeping research accountable, transparent, and reproducible.

Experience

Work history, roles, and key accomplishments

Current

AI for Neural Laser Stimulation

Current

Autonomous University of Madrid

Jun 2026 - Present (1 month)

Apply and evaluate transformer-based and foundation-model approaches for time-series, with emphasis on robustness, interpretability, and physiological plausibility. Develop ML workflows for in vitro neural recordings to characterize stimulation-response and build classifiers against predefined validation criteria.

Foundation Models Robustness Model Interpretability Representation Learning Physiological Plausibility In Vitro Neural Recordings response evaluation Classifier Design

Growth Strategy Lead

Silk Cashback

Mar 2026 - Jul 2026 (4 months)

Designed and analyzed A/B experiments and performed SQL/BigQuery analyses of user behavior. Translated results into recommendations for non-technical stakeholders.

A B Testing SQL BigQuery User Behavior Analysis Stakeholder Recommendations

ML for Biosignals

Autonomous University of Madrid

Apr 2025 - Mar 2026 (11 months)

Developed an end-to-end ML workflow for clinical dementia EEG datasets, including entropy/complexity feature extraction and stability selection. Used nested and leave-one-subject-out cross-validation with permutation testing and surrogate analysis to validate that effects reflected signal rather than confounds, producing interpretable feature-level explanations of classifier behavior.

EEG Stability Selection Cross Validation Permutation Testing Surrogate Analysis Error Confound Validation

ML & Neural Signal Processing

Autonomous University of Madrid

Apr 2021 - Mar 2025 (3 years 11 months)

Designed and evaluated a self-attentive BiLSTM architecture for temporal decoding from multi-modal EEG/EMG time-series with subject-wise validation. Built the open-source ExSEnt entropy method and released reproducible analysis pipelines covering preprocessing, feature extraction, source localization, and statistical modeling.

Self Attentive BiLSTM Temporal Decoding Attention Mechanisms Hyperparameter Tuning Slurm ExSEnt EEG Preprocessing Feature Extraction Localization Statistical Modeling EEGLAB Extension

Visiting Researcher

Swartz Center for Computational Neuroscience

Apr 2023 - Jun 2023 (2 months)

Advanced electrophysiological source imaging by performing ICA component selection, dipole fitting, and group-level clustering. Worked with EEGLAB/DIPFIT and custom MATLAB pipelines.

Electrophysiological Source Imaging Dipole Fitting Clustering EEGLAB DIPFIT Matlab

Computational Modelling of Cognition

Amirkabir University of Technology

Oct 2013 - Apr 2018 (4 years 6 months)

Modeled brain-state transitions between coma and consciousness using nonlinear dynamical systems. Published results in a Q1 ISI journal.

Dynamical Systems Brain State Transition Modeling

Education

Degrees, certifications, and relevant coursework

Autonomous University of Madrid

Doctor of Philosophy (Ph.D.), Computer Science

2020 - 2026

Grade: Cum Laude

Ph.D. in Computer Science (Cum Laude) focusing on characterizing biosignal dynamics through spectral and complexity features for motor activity and dementia classification (EEG/EMG).