Startup Funding Intelligence & Market Dashboard — Repository

Built an NLP-based sector classification model on Databricks using Apache Spark, automating startup pattern recognition and reducing manual deal-triage time by an estimated 70%. Designed an end-to-end ML pipeline with dbt transformations and LLM-based feature extraction, generating a data-driven VC investment thesis that identified underserved market sectors. Enforced Pydantic schema validation on all LLM outputs before downstream consumption, ensuring model reliability and preventing corrupted predictions from propagating through the analytics pipeline.

Patient Readmission Analytics & Prediction System — Repository

Built an ML pipeline to predict 30-day hospital readmission risk from 70K diabetic patient encounters, aiming to help flag high-risk patients before discharge. Handled severe 11:89 class imbalance using SMOTE and cost-sensitive threshold tuning rather than optimizing for accuracy. Benchmarked Logistic Regression, Random Forest, and XGBoost with MLflow experiment tracking; selected the final model using a custom clinical cost function, achieving 0.65 AUC-ROC and 99% recall on high-risk patients.

IIT International Student Policy Chatbot — Repository

Achieved 90.4% accuracy on a 52-question benchmark, outperforming AutoRAG and NotebookLM baselines deployed live 24/7 with real IIT students in closed beta. Designed a Neon PostgreSQL feedback analytics layer capturing every user session and query, enabling continuous performance monitoring and iterative improvement post-launch. Built hybrid BM25 + vector search with RRF fusion across 328 policy document chunks, improving retrieval precision across both keyword and semantic query types.

✅ Verified by Extern · ⏱️ In progress

Excel & AI Financial-Modeling Sprint

Build a three-statement model, forecast with drivers, run a DCF, and pitch a company like a pro. In this 3-week sprint, GPT helps you work faster—so you can walk away with real valuation skills and a polished investor deck.