Built a full-stack internal auditing chatbot and platform from the ground up, enabling 200+ auditors to verify insurance claims over $25,000 during the prepay step before Aetna processed payouts, improving overall auditing efficiency.
Engineered a BigQuery and SQL data layer paired with a TypeScript frontend and Python FastAPI backend, deploying the platform into production.
Designed an orchestrator agent with 3 sub-agents powered by Gemini 3.5 Flash to integrate 3 new tools into a unified auditor workflow.
Presented the platform to business leaders on multiple occasions, driving adoption as the sole auditing tool used by a team of 200+ auditors.
BigQuerySQLTypeScriptFastAPIGeminiAI Agents
Inmobi
Machine Learning Engineering Intern · San Mateo, CA
Apr 2025 – Aug 2025
Developed and deployed a production-grade predictive pricing model to predict minimum-bid-to-win, leveraging 10+ large-scale signals, increasing auction efficiency and revenue outcomes.
Engineered 3+ scalable data pipelines for model training and custom feature creation, ensuring robustness and performance in a high-throughput production environment.
Authored a comprehensive data dictionary for 100+ key features and metrics, enhancing cross-functional collaboration and accelerating model-driven product development.
Machine LearningData PipelinesPredictive ModelingProduction Systems
StudyPlanGPT
Founder · Boston, MA
Sep 2024 – Apr 2025
Created a personalized study planning app leveraging 2+ GPT models to generate tailored study recommendations, improving study efficiency and time management.
Launched the app on the Apple App Store to help students create dynamic, AI-driven study plans.
Implemented NLP-based recommendation algorithms that adapt plans to user behavior and preferences, driving engagement for 1000+ users.
Built a Flask web app that prices dynasty fantasy football trades with live FantasyCalc data and finds balanced trades across every roster in a Sleeper league, adjusted for league format and team strategy.
Built an offline data science pipeline with aging curves, a gradient-boosted next-season production model (MAE ~42 vs 47 baseline over 7 walk-forward seasons), and a "buy low / sell high" market signal.
Calibrated the fairness model on 280 real accepted trades using grid search and bootstrap intervals, with a stateless, tested, production-ready deployment.