Nikita Patil

Roadblock Title:

Building Private, Low-Latency GenAI on Mobile for Fintech Apps

Time:

Tuesday - 11:30 AM (Tower D)

Abstract:

Shipping GenAI in fintech mobile apps hits a hard roadblock: cloud-only inference creates latency, cost, and privacy/compliance risk—while on-device inference can crash, overheat, or miss accuracy targets if you treat it like backend ML. In this talk, I’ll share a production-minded playbook for building “tiny giants” using small language models on iOS/Android (Core ML, TensorFlow Lite, MediaPipe). We’ll walk through quantization and performance tradeoffs, a repeatable resource-budget checklist (memory/thermal/battery), and a hybrid routing template that switches edge vs cloud based on network, privacy class, and cost. Includes redaction and fallback patterns for sensitive data.

Bio:

Nikita Patil is a Senior Software Engineer at Intuit, where she builds the QuickBooks Workforce mobile app—helping small businesses manage time tracking and payroll from their phones. She specializes in iOS platform engineering and previously developed an infrastructure framework that enables teams across Intuit to produce, consume, and contribute to the company’s product ecosystem at scale. Today, she’s focused on pushing the boundaries of mobile by exploring how on-device AI can make apps faster, smarter, and more private.

Outside of work, Nikita mentors junior engineers, enjoys Zentangle sketching, and is a proud mom to her “cute little pie.