Vridhi
Wadhawan

Markets signal before they break. Built a multi-method detection pipeline — then validated every flagged anomaly against real financial news to test whether the signal actually meant something.

LLMs hallucinate when they don't know. Built a retrieval pipeline grounded in Wikipedia sources on Indian Prime Ministers — baseline accuracy of 0.19 improved to 0.37 through pipeline design alone.

Data without a decision interface is just storage. Built an executive dashboard that translates KPIs into the exact view a decision requires.

A model that isn't deployed isn't a product — it's a notebook. Built the full pipeline from training to retraining trigger so the system stays live.

The optimal product was already known. The harder question was who actually values it — and what that means for portfolio strategy. Affinity-based segmentation to find the answer.

Built a live forecasting app for any NSE stock — XGBoost predictions, Gemini-powered narrative insights, drift monitoring, and R² health checks. The question was never just accuracy — it was whether the system could tell you when to stop trusting it.

Relational databases store entities. Knowledge graphs store relationships — and relationships are where the real questions live. Designed the ontology for a residential complex as a digital twin.

Four case studies across four data modalities — tabular, text, image, and sequential. Each one chosen to build intuition for when an architecture earns its complexity.