Systems Administrator · Denison University · Granville, OH
AS Abhishek Shekhar

Applied ML · Computer Vision · NLP · Health

Abhishek Shekhar

Systems Administrator, Denison University · Applied ML Researcher

I work on small, careful machine learning systems — image captioning, clinical text compression, climate pattern classification — and on the surrounding craft of making them reproducible, evaluable, and useful. My projects sit at the seam between modern deep learning and the operational reality of getting models into the hands of people who need them.

Explore my research See projects Get in touch
Portrait of Abhishek Shekhar

Recent

News & updates

All updates →
May 2026

MedCompress: compression-as-clinical-context

Building MedCompress, a tool for collapsing long medical documents into structured, faithful summaries usable by downstream NLP pipelines and clinical decision-support workflows.

2023

Climate change classification with SOM, ANN, and CNN

Unsupervised regional clustering of atmospheric data followed by supervised classification of climate-impacted regions, evaluated with F1 and confusion matrices on global maps.

Dec 2023

Graduated Summa Cum Laude from Beloit College

BSc in Computer Science, GPA 3.87, Phi Beta Kappa. Presented Image Caption Generator (CNN + LSTM, CUDA-tuned) at the 47th Annual Beloit Student Symposium.

2023

RNN vs. CNN on clinical tabular data

Comparative study of recurrent and convolutional models on the UCI Heart Failure Clinical Records dataset, with K-Means and SOM clustering used to surface patient groupings.

Focus

Research & engineering areas

Where my IT work and ML work talk to each other.

Read more →
Applied deep learningVision, NLP, and sequence models on real datasets
Computer visionCNN encoders, image captioning, feature extraction
Sequence modelsRNN, LSTM, and CNN architectures compared
NLP & text classificationTF-IDF baselines, neural pipelines, evaluation
ML for healthClinical tabular data, MedCompress, decision support
ML for climateAtmospheric clustering and regional classification
Unsupervised methodsSOM, K-Means, dimensionality reduction
Reproducible evaluationBLEU, F1, AUC, confusion matrices, ablations

Featured

Selected projects

All projects →
Image Caption Generator

CNN + LSTM image captioning, CUDA-tuned

End-to-end image captioning pipeline using VGG16 as a CNN encoder paired with an LSTM decoder, trained on Flickr 30k. Optimized CUDA kernel configuration for parallel batch processing, evaluated with BLEU, presented at the 47th Annual Beloit Student Symposium.

PyTorch TensorFlow CUDA LSTM VGG16
MedCompress

Compression as clinical context

A tool for collapsing long medical documents into compact, structured summaries that preserve the clinical signal needed by downstream pipelines and decision support. Built to slot into existing EHR-adjacent workflows rather than replace them.

NLP Health LLMs Decision support
Climate Change Analysis

Unsupervised + supervised pipeline for atmospheric data

Trained Self-Organizing Maps on surface temperature, precipitation, and pressure readings to cluster geographic regions, then layered ANN and CNN classifiers to identify areas with the strongest climate change signal. Reported with confusion matrices and regional visualizations on global maps.

PyTorch SOM ANN CNN Climate
Heart Failure Classification

RNN vs. CNN on clinical tabular data

Binary classification on the UCI Heart Failure Clinical Records dataset, predicting mortality risk from clinical features. K-Means and SOM clustering surfaced natural patient groupings before supervised training. RNN and CNN performance compared with confusion matrices, precision, recall, and AUC.

PyTorch RNN CNN K-Means SOM

Want to collaborate or chat?

Always happy to talk about applied ML, computer vision, sequence models, or ML for health and climate.

Say hello