Shreyas Sabnis

• Actively involved in development and tuning of NAVI, Coupa’s LLM‑powered multi‑agent AI Assistant • Built robust evaluation frameworks, using self‑reflection to track cost, latency and accuracy metrics for three agents • Applied advanced prompt engineering techniques and introduced hydrid search to improve performance metrics • Improved Accuracy: Router Agent from 82% → 92%, Forms Agent from 62% → 78%, Data Table Agent from 63% → 84% • Expanded the Data Table Agent to support 12 new Data Tables, increasing the total supported table count from 30 to 42 • Currently developing the Smart Intake Orchestration Agent to guide users through the intake process

Quantitative Research Intern • Engineered versatile order fill simulator for medium frequency trading strategies, utilizing immediate future data • Incorporated order-book VWAP percentiles, impact cost maps, and historical fills in decision-making of each fill • Conceptualized novel statistical indicators using full-day order book and trades data of entire option chains • Created an efficient automated pipeline to generate those stats at end-of-day in C++ via in-house packages

Modem Systems Wireless RnD • Used TRACE32 emulator to simulate Snapdragon Radio’s communication links, producing crucial crashdump data • Created a Python tool for analysis of crashdump report, for pinpointing origins of excessive power consumption • Deployed a script on Jenkins to generate a tabber-style HTML report providing front-end with a user-friendly UI