HrishikeshTavar

A four-agent AI system automating a real sales-strategy planning task, architected against silent arithmetic errors and schema drift — with a self-correction loop and 22-criterion rubric scoring.

100/100 rubric score

Ten independent applied-AI tasks spanning classical ML, deep learning, NLP, tabular foundation models, and LLM prompt-engineering — built for breadth across the current applied-AI toolkit.

Unsupervised risk-tier segmentation (PCA + K-Means) on 50,000 loan applications, paired with a PyTorch default-prediction model evaluated by risk segment rather than in aggregate.

An OLTP-to-star-schema data warehouse for a wholesale food distributor, with synthetic data calibrated against real UK food-industry benchmarks and Tableau dashboards on margin, wastage, and seasonal demand.

A binary classifier identifying which bank customers are likely to apply for credit, enabling proactive retention outreach in place of reactive processing.

0.85 AUC · 0.73 Recall

Two independent regression case studies — what drives household food spending in England, and what drives daily podcast listenership — testing when interaction effects are actually justified by the data.

A statistical analysis of Underground carriage temperatures (2013–2024) — outlier detection, Welch's t-test, and effect sizes — to inform where to pilot new air-cooling infrastructure.

Exploratory analysis correlating COVID-19 case data with the World Happiness Report across countries.