Hello, I'm Atul Kumar

Built an end-to-end instance segmentation pipeline to detect dental caries from radiographic images using YOLOv8s-Seg (PyTorch). Trained the model on high-resolution images (768×768) using AdamW optimizer and cosine learning-rate scheduling, prioritizing recall to reduce false negatives in a medical setting. Evaluated performance using Precision, Recall, mAP@50, and mAP@50–95, achieving 22.6% mask recall on the validation set.

Analyzed weekly NIFTY 50 data (2018–2025) using rolling 5-week returns and volatility to characterize market behavior.Applied unsupervised clustering to classify market regimes into bull, bear, and sideways states based on return–volatility patterns. Identified bull regimes with returns 0.02–0.10 and volatility below 0.03, while bear regimes showed drawdowns beyond −25% with volatility up to 0.10.Evaluated regime-based strategies, demonstrating momentum buy-and-hold effectiveness in bull markets while limiting losses during bear phases.

Analyzed 7 large-cap equities using daily price data (2022–2025) to evaluate performance across varying market conditions.Computed key risk metrics including annualized volatility (2.42–22.28), Sharpe ratio (20.48–94.72), and maximum drawdown (−30% to −74%).Identified AAPL as the strongest risk-adjusted performer (Sharpe 94.72) while flagging high-risk stocks such as NVDA (volatility 22.28).