Adarsh Kumar

A GAN pitted generator against discriminator over 10k+ WikiArt paintings until its portraits blend into the source gallery.

DenseNet121 classifier for 7 skin-disease classes across 9,000+ dermoscopic images, crossing 90% accuracy with heavy augmentation.

A deployed Streamlit app that forecasts the next 5 trading days for any Yahoo Finance ticker with an LSTM trained on historical price data.

Parses uploaded resumes, classifies them by skill domain with an NLP pipeline, and ranks candidates against a job description for first-round screening.

CNNs over ASL digit key-points and ISL alphabet images, with EDA across both sign languages — reaching 100% classification accuracy on the digit set.

Flask app that predicts AQI for Indian cities from scraped real-time pollutant data — benchmarked 10+ models (XGBoost, LightGBM, CatBoost) and serves health advisories.

Season 6 Playground builds — EV interest, F1 pit stops, irrigation need — logged over 140 experiments of stacking, pseudo-labelling and knowledge distillation.

Identifies bird and frog species from Pantanal wetland soundscapes — linear probes trained on Perch v2 embeddings, iterated through six logged model versions.

Two CNN models estimating age (0–116) and gender from 20,000+ UTKFace faces, shipped as reusable .h5 models with a combined prediction notebook.

Neural text classifier over 40k+ labelled real vs fake articles, with word-cloud EDA and full precision, recall and confusion-matrix evaluation.