SHIPPED,NOT SHELVED.

Ten specialized agents run in sequence and in parallel: safety screening, entity extraction, RAG over verified clinical guidelines, medical reasoning, dialogue tracking, response composition, and deliverable generation, with both text and video (Tavus) pathways. It reaches 99.5% emergency detection accuracy on a safety-critical test suite at under 2 seconds end to end, with a 40% latency cut from parallel execution across 10 fault-tolerant modules. A fine-tuned LLM simplifies clinical jargon for patients at a 92% SARI score. Built with Next.js, FastAPI, WebSockets, and RAG via Ragie, deployed on Vercel and Render.

A 5-architecture meta-ensemble (CNN, DFNN, Transformer, Random Forest, XGBoost) that resolves domain shift across 11 geological formations. Physics-informed features (Eaton ratio, pressure gradients, normalized depth) plus SHAP and LIME for interpretability. On 281,000+ samples from 21 wells it reaches R squared of 0.92 on blind test wells, 27% over the best single model, at under 50ms inference, with strict well-based splitting so there is no leakage. The research behind the IJCNN 2026 paper.

Deep learning segmentation of brain tumors from multi-modal MRI (T1, T1CE, T2, FLAIR) using a U-Net architecture, reaching a 94% Dice coefficient across tumor regions.

A four-phase CLI (discovery, enrichment, processing, output) that unifies Semantic Scholar, OpenAlex, and GitHub to find and catalog every paper citing NDIF. It deduplicates across arXiv IDs, DOIs, and fuzzy title matching, runs an LLM classifier for paper context, and extracts figures from PDFs concurrently. It replaced a manual workflow and cut effort by 96%.

Predicts loan default risk using ensemble methods (XGBoost, scikit-learn) with feature engineering across borrower profiles. 89% ROC-AUC.

A web app that detects and identifies multiple celebrity faces in an image using OpenCV and a trained classifier, served through Flask. 87% accuracy.

Predicts home prices across 240+ Bangalore locations with a regression model behind a Flask web app. 82% accuracy.

A complete email-campaign platform with OAuth authentication, Stripe payments, and campaign analytics, built end to end on the MERN stack.