Ebrahim Jahanshiri
AI/ML engineer
Hi, I’m EJ!
I build machine-learning systems: language models that read documents and justify a score, deep-learning models on satellite imagery, and models that recover a heart rhythm from EEG.
My journey in data science began with a Pentium IV computer in 1995 that allowed me to work with earliest versions of Microsoft Excel and Fortran!. I quickly learned about how to load data, create charts and even develop out-of-the-box linear regression models with some model diagnostics. I carried all these skills with me throughout my studies in the university where I worked on various projects that required complex 3D geospatial data manipulation, analysis and publications. I have mastered R in my PhD, where I developed econometrics models for prediction of house property values.
All I have learned by myself were not enough, so I have recently completed a M.S. in Data Science (a joint program with Gothenburg and Chalmers universities, Sweden). This strong academic foundation, coupled with practical experience, has allowed me to hone my skills in areas like Machine Learning, Predictive Modelling, Data Visualization, and Big Data Analytics. Over the years, I’ve had the privilege of working in diverse industry settings. My quest for unraveling the hidden patterns in data has led to successfully enhancing business operations, improving customer experiences, and predicting market trends.
My technical toolkit is Python, R, and SQL, with recent work in large language models, geospatial deep learning, and machine learning on EEG and ECG. I use PyTorch, TensorFlow, and cloud deployment on GCP. A believer in constant learning, I’m always seeking to update my skills and stay abreast of the latest industry advancements.
But beyond the algorithms and code, my true passion lies in storytelling. Data, in my eyes, is a narrative waiting to be discovered and shared. I aim to demystify the complexities of data and present it in a way that is not only comprehensible but also actionable for decision-makers.When I’m not knee-deep in data, I enjoy hiking, reading about cognitive science, and exploring the local coffee scene. I’m also an avid supporter of women in tech, dedicating part of my time to mentoring aspiring female data scientists. Welcome to my little corner on the web, where I share my projects, thoughts on data science trends, and resources for fellow data enthusiasts. I’m excited to connect, collaborate, and keep pushing the boundaries of what data can achieve.
Thank you for stopping by!
- Large language models
- Machine learning for medical signals
- Geospatial deep learning
- MLOps and automation
- Storytelling with Data
- Personal development
MSc Applied Data Science, 2022
Gothenburg and Chalmers Univeristies
PhD GIS and Geomatics Engineering, 2013
Universiti Putra Malaysia
MSc Precision Farming, 2008
Universiti Putra Malaysia
Bsc Engineering, 2000
Ferdowsi University of Mashhad
Skills
Retrieval-augmented generation, LLM-as-judge scoring, document understanding, criteria extraction
PyTorch, TensorFlow, detection models, generative models
EEG and ECG, independent component analysis, time-series classification, quality scoring
SAR and optical imagery, Google Earth Engine, detection and mapping
NumPy, Pandas, Scikit-learn, PyTorch, TensorFlow
Docker, Cloud Run, GCP, DVC, MLflow
Inference, hypothesis testing, model validation, non-parametric methods
PostgreSQL, MySQL, Neo4j, GraphDB
Experience
- Machine learning for medical-device signals, starting with HeartBrain: recovering heart rhythm from multichannel EEG when no ECG channel is available.
- ECG-free heartbeat detection with independent component analysis and a quality score. On held-out recordings the score selected beats with median F1 0.94 and heart-rate error of 0.4 bpm.
- Atrial-fibrillation screening from EEG-derived beats (AUC 0.985 under typical EEG error) and heartbeat-evoked potentials in autistic and typically developing children.
- Preliminary results from November 2026. Planned outputs are a methods paper, an atrial-fibrillation-from-sleep-EEG paper, and an open-source Python toolbox.
- Chief technology officer and hands-on AI lead for Nexus, a maritime Earth-observation platform, taking the system from TRL 5 toward validated TRL 6/7.
- Led multi-sensor ingestion and preprocessing for SAR and optical imagery, including radiometric and geometric correction and dataset versioning with DVC and Git.
- Trained and validated detection models in PyTorch and TensorFlow, reporting precision, recall, F1, and confusion matrices, and tested cloud deployment.
- Co-led inference APIs (JSON and GeoJSON) and verification of the WebGIS dashboard used to review detections.
- Designed and built an evidence-based candidate shortlisting service that ranks applicants from a job description and résumés: https://shortlisting-1016040926675.us-central1.run.app
- LLM pipeline for document ingestion, explicit and inferred hiring criteria, retrieval-augmented enrichment, and LLM-as-judge scoring with a written justification per criterion.
- Rankings are decision support for a recruiter, with checks for hallucination, bias, and explainability. Deployed on Google Cloud Run.
- I design, develop, and deliver data and AI products for the EU RADIANT project.
- I use Generative AI (LLMs, Diffusion and Adversarial models) to develop apps and website deployed on GCP, AWS.
- Built a database of more than 500 European agronomic studies (1972–2022) and a knowledge base for agricultural diversification on GCP.
- Shipped AiSQLquery, a retrieval-augmented extension for domain questions, and an MLOps pipeline with MLflow and DVC.
- Led the data work for the RADIANT Horizon 2020 project (220,000 euro) and a UK diversification analysis across climate, soil, and markets.
- Built suitability workflows on SQL, Google Earth Engine, and GIS, including a national 3D soil database for Sri Lanka.
- Used spatial GANs and other models to estimate soil properties where samples were missing.
- Thesis: https://gupea.ub.gu.se/handle/2077/72200?show=full
- Led 24 specialists building the CropBASE apps, under a 290,000 euro Horizon 2020 grant for a land decision-support system.
- Applied support vector machines to economic feasibility and published the evidence base.
- Built software used by more than 10,000 farmers in Southeast Asia, including a nutrition-security tool and a crop-model input generator.
- Compared spatial regression models for property valuation and built PFMap, an ArcGIS tool for precision-farming maps.
- Co-developed hyperspectral analysis of oil-palm fruit that led to two patents.
- Fellowship from the Malaysian Centre of Remote Sensing. A geostatistical mapping method cut the samples needed by about 70 percent.
- Segmented Landsat imagery and held a national grant for a cellular-automata planning system.
- Built a metadata system for remote-sensing data fusion in a Southeast Asia land-cover project.
- Transcribed records and searched the literature for related land-cover studies.
Selected projects
HeartBrain
ECG-free heart-rhythm recovery from multichannel EEG, for atrial-fibrillation screening and heart–brain measures.
Nexus
Maritime Earth-observation platform for multi-sensor detection, from data pipelines through validated deep-learning models.
Evidence-based LLM shortlisting
A Cloud Run service that ranks candidates from a job description and résumés, with a written justification for every score.
Gallery
Featured Publications
Contact
Drop a line to connect or alternatively schedule a meeting: