At CRED, I built conversion-probability and ticket-size models that feed a Gurobi optimization program for allocating lender offers. The system increased monthly disbursal by ~INR 50Cr and saved ~INR 10Cr/year in lender penalty payouts.
I also built databricks-mcp, a production platform giving AI agents access to Databricks, and the flagship data-science agent on it. It supports 51 agents and 400+ users across 50 teams, with 99.7% reliability.
For CRED’s vehicle business, I built a vehicle make-model-variant mapping model using CatBoost and a fine-tuned SentenceTransformer. After a staged A/B rollout, it reduced support tickets by 90% and generated a premium upside of ~INR 8Cr/year.
As a Data Scientist Intern at CRED, I developed an XGBoost name-validation model that improved F1 from 0.90 to 0.98 and was adopted company-wide. I also developed a motor-risk model that drove a 10% uplift in user conversion and a 5% uplift in premium overall.
