← J·A — homeindex / portfolio
project index

PORTFOLIO

Six selected builds — streaming pipelines, medical-imaging models, production AI and full-stack platforms. Click any card for the full write-up.

P·01read more →

Moby Dublin — Spark telemetry

Event-time-windowed Spark Structured Streaming pipeline with fault-tolerant aggregation for a distributed fleet telemetry system, surfacing demand hotspots and crowding signals.

Apache SparkAWS S3MongoDBParquet
duplicate GPS records−14%
P·02read more →

Alzheimer's disease detection

Interpretable Multi-IT Transformer with Triplet Attention staging Alzheimer's across 4 classes on 44,000 MRI scans, benchmarked against 4 CNN baselines on an NVIDIA A100.

PyTorchTransformersTriplet AttentionA100
MRI scans processed44,000
P·03read more →

Skin cancer classification

CNN models for skin lesion classification in TensorFlow and Keras — InceptionV3, MobileNetV2 and VGG variants with augmentation, reaching 86% accuracy.

TensorFlowKerasInceptionV3MobileNetV2
classification accuracy86%
P·04read more →

AI support chat — Amazon Bedrock

AI-powered customer support chat at Easebuzz, trained on internal product policy and deployed across merchant support operations to deflect repetitive queries.

Amazon BedrockPythonDjangoLLM
environmentproduction
P·05read more →

Data mining & ML

End-to-end analysis: cleaning, feature engineering and EDA, then supervised models in scikit-learn compared across evaluation metrics.

scikit-learnPandasEDAFeature engineering
models comparedmulti-metric
P·06read more →

Akshaya Patra fleet tracker

Intern project tracking food vans dispatched from kitchens to schools — driver and route details, built in Java with Git.

JavaGitLogisticsSocial impact
built duringinternship

More experiments, coursework and spikes live on GitHub.

github.com/JosephJ7 ↗