Mathematics · Data Science · Environmental Research

Ajin Thomas.

Computational research
with an environmental direction.

I am a Data Science graduate with a background in Mathematics and research experience in scientific machine learning, numerical modelling and multivariate time series. I am currently investigating generative models for rehabilitation robotics at Scuola Superiore Sant’Anna in Pisa.

My long-term aim is to apply these methods to geography and Earth system science, understanding how human activities and natural processes shape landscapes and ecosystems.

Explore my research ↓Get in touch ↗

GitHub LinkedIn Academic background ↓

Academic trainingM.Sc. Data ScienceFAU Erlangen–Nürnberg

Mathematical foundationB.Sc. MathematicsUniversity of Kerala

Current researchScuola Superiore Sant’AnnaGenerative modelling · Pisa

01 / Research

Selected investigations.

Three projects that explain my research direction, from scientific machine learning to the environmental questions I want to pursue.

01 / Geospatial

Independent environmental investigation

The Earth Beneath the Cloud

Investigating how data-centre development may affect vegetation, water resources, drainage and nearby ecosystems. Starting with documented site boundaries in Visakhapatnam and preliminary satellite vegetation analysis; expanding toward multi-site comparison.

Research status

Documented site boundaries and preliminary NDVI analysis inform the investigation. Seasonal comparison, independent water evidence and defensible controls are needed before causal claims.

Satellite observationsGISNDVIEnvironmental research

Interactive explanation ↗

02 / Scientific ML

Master’s thesis / research

Physics-informed lithium-ion battery health prediction

Developed a hybrid PyTorch model for battery state-of-health estimation using 16 charging-curve features from experimental cycling data (55 cells, six protocols). Combined learned dynamics, an automatic-differentiation residual and monotonicity constraints; compared MLP/CNN baselines and cross-dataset transfer.

Reported thesis result

Best reported configuration (batch 6; four training batteries; mean over 10 runs): MAE 0.0089, RMSE 0.0119, MAPE 0.97%. Thesis Table 5.2.

PyTorchPhysics-informed MLModel benchmarkingTransfer learning

Source code Interactive explanation ↗

03 / Scientific ML

Research internship · ongoing

Rehabilitation robotics signal generation

Investigating conditional diffusion models for multichannel position and force signals. Using EDM, waveform and wavelet representations, sparse clinical score conditioning, sampler and denoiser ablations, and patient-disjoint validation.

Research status

Ongoing research. No patient recordings, confidential dataset details or unpublished results are shared.

Generative modelsTime seriesPatient-disjoint validation

Interactive explanation ↗

The illustrations above are conceptual diagrams, not satellite imagery, patient signals, or measured model outputs.

Research direction

Understanding the natural world through computation.

My academic training has been in mathematics and data science. My research interests have increasingly moved toward geography and environmental change. I want to develop methods that help us investigate the Earth using reliable observations, mathematical models and careful evaluation.

01Landscape & ecosystem changeEarth observation, forest dynamics, land-use change and the carbon cycle.

02Water & environmental systemsHydrology, drainage, land use and the ecological effects of infrastructure.

03Scientific modellingData-driven and physics-informed methods, uncertainty and validation.

These are research interests and learning goals; not claims of specialist expertise in ecology or hydrology.

02 / Project archive

More computational work.

Selected scientific, geospatial and applied AI projects. Public code is linked where available.

8 additional projects

EngineeringDec 2025–Feb 2026

Wind forecasting for grid balancing

Developed 24-hour-ahead forecasts using five years of hourly meteorological data in four German grid regions; compared LSTM, Temporal Fusion Transformer, XGBoost and LightGBM, including quantile predictions, with API and containerised workflows.

LSTM / TFTXGBoostFastAPIDockerTime series

Public code repository; forecasts and evaluation details should be checked in the project documentation.

Source code

Scientific MLFeb–May 2026

Fourier Neural Operator · Navier–Stokes

Implemented a PyTorch Fourier Neural Operator for 2D vorticity evolution using pseudo-spectral simulation data; assessed cross-resolution generalisation and numerical-model inference.

PyTorchNeural operatorsNumerical simulationPDEs

Reported relative L2 error ≈0.025 at 64×64; Special Award for Innovative Projects at the 4th EELISA Scientific Student Competition (as recorded in my CV).

GeospatialJun–Sep 2025

Trek Knowledge Project

Developed trail-network routing and geographic analysis with OpenStreetMap, OSMnx, GeoPandas and NetworkX; integrated a Llama-based model via Groq for contextual route information and questions, using FastAPI and Leaflet.

OSMnxGeoPandasNetworkXLlama / GroqFastAPI

Public repository documents route analysis and LLM integration. This is a technical project, not a deployed product or autonomous GIS agent.

Source code

EngineeringApr–Jun 2025

Natural-language data interaction platform

Built a Python, Streamlit and LangChain interface using the OpenAI API to support natural-language exploration and analytical summaries of more than 100,000 structured records.

LangChainOpenAI APIStreamlitPandas

Project summary; public source code is not linked.

Scientific MLSelected technical project

Physics-informed inverse heat transfer

Explores inverse heat-transfer questions through physics-informed neural modelling. The public repository is the authoritative source for implementation details.

PINNsHeat equationInverse problems

Public code available; no benchmark results are reported here.

Source code

EngineeringSep 2026

Industrial seal identification · computer vision

Built a YOLO26 Nano digit-detection pipeline with annotated data and compared its recognition behavior against an Ollama vision model under different image conditions.

Computer visionObject detectionYOLOOllama

Hackathon prototype; no unverified recognition accuracy or latency numbers quoted.

EngineeringMay–Jun 2026

Customer retention and revenue analytics

Designed SQL analytical views and Power BI summaries for public e-commerce data, exploring retention, customer segmentation, revenue trends and geographic patterns.

PostgreSQLSQLPower BIKPI analysis

Project summary; dashboard not publicly linked.

EngineeringGitHub project

Stock-price ensemble prediction

Repository investigating NVDA stock-price forecasting using an ensemble machine-learning approach.

Time seriesEnsemble MLPython

Public repository; results are not independently evaluated here.

Source code

03 / Academic & professional journey

Experience and education.

My foundation in mathematics, computational research and practical data analysis.

Research & professional experience

Sep 2026–present

Research Intern · Generative Modelling and Sensor Data

Scuola Superiore Sant’Anna · Pisa, Italy

Developing conditional diffusion models (EDM) in PyTorch for position and force time series, using wavelet time–frequency and waveform representations with sparse Fugl–Meyer Assessment conditioning. Evaluating temporal structure, cross-channel dependencies, conditioning sensitivity and potential augmentation utility using controlled ablations and patient-disjoint validation.

PyTorchEDMMultivariate time seriesPatient-disjoint validation

Oct 2021–Sep 2022

Data Analyst Intern

Infosys Ltd. · Bengaluru, India

Built Python/Pandas/NumPy and SQL preprocessing, validation and reporting pipelines for large datasets, including million-record-scale data; built Power BI dashboards and evaluated classification models.

PythonSQLETLPower BIScikit-learn

Education

2022–2026

M.Sc. Data Science

Friedrich-Alexander-Universität Erlangen–Nürnberg (FAU), Germany

120 ECTS; final grade 2.0 (German scale). Focus on machine learning, simulation and numerics, and statistical data analysis.

Deep learningExplainable machine learningMachine learning for time seriesPattern recognitionNumerical methods

2018–2021

B.Sc. Mathematics

Mar Ivanios College, University of Kerala, India

120 Indian CBCS credits; CGPA 9.13/10 (A+). Core mathematics (52 credits), physics and statistics (14 credits each), and astronomy and astrophysics.

Linear algebraDifferential equationsStatisticsIntegral transformsPhysics

04 / Methods & tools

Technical toolkit.

Methods and technologies I have worked with in coursework, research and technical projects. Geospatial and Earth observation skills are an active area of development.

Programming & scientific computing

PythonNumPyPandasSciPySQL / PostgreSQLGit

Machine learning & numerical modelling

PyTorchTensorFlowScikit-learnXGBoostLightGBMCNNsLSTM / TransformersPINNsFourier neural operatorsEDM diffusionModel validation

Geospatial & Earth observation

GeoPandasOSMnxNetworkXOpenStreetMapKML / GeoJSONGoogle Earth Engine (initial use)Preliminary NDVILeaflet

Applied AI & engineering

LLMs / RAGLangChainOpenAI APIGroq APIOllamaYOLOFastAPIDocker / ComposeStreamlitn8n

Mathematics & visualisation

Linear algebraDifferential equationsStatistical inferenceNumerical simulationMatplotlibPlotlyPower BI

05 / Recognition & learning

Awards, training and certifications.

Special Award for Innovative Projects

4th EELISA Scientific Student Competition · Madrid · 2026

Erasmus+ mobility support

Research collaboration with Scuola Superiore Sant’Anna · 2026

Continuing education

ISRO–IIRS · AI/ML for Geodata Analytics · 10-hour participation · 2026
NASA Open Science Essentials · training certificate
Bavarian–Czech AI Summer School · AI and Industry · 2026
Building with the Claude API · Anthropic · 2026
Python for Data Science, AI & Development · IBM
BCG GenAI Job Simulation · Forage · 2025
Deloitte Job Simulation · Forage

06 / About

The person behind the research.

I grew up in Kerala, India, and studied mathematics before moving to Germany for my Master’s degree in Data Science. My growing interest in geography comes from curiosity about natural environments and the ways they change.

Outside research, I enjoy alpine hiking and mountaineering, climbing, cycling, reading and drawing. Exploring different landscapes keeps my interest in the natural world grounded in everyday experience.

Languages: Malayalam (native) · English (C1) · German (B1)

07 / Contact

Let’s discuss research.

I welcome conversations about doctoral research, scientific computing, environmental data science and research-oriented technical collaborations.

ajinalummottil098@gmail.com ↗GitHub LinkedIn