Deva Shriyansh Annamaraju

Software Engineer 2 (promoted Apr 2026)

• Architected and scaled a React Native workforce management app from scratch to 150K+ monthly active users, 1M+ Google Play downloads and 99.9% crash-free sessions, with offline-first flows, shared UI primitives, feature flags and native Android and iOS integrations.
• Cut bundle size and made cold start 40% faster; built foreground service helpers and reliability tooling for high-frequency field use.
• Built an on-device face recognition pipeline on Google ML Kit with custom embedding logic, replacing a vendor SDK, saving ₹10 million a year and verifying in under 200 ms on mid-range devices.
• Own mobile release engineering on Google Play and the App Store: requirements shaping, GitHub Actions CI/CD, production rollouts and release quality monitoring.
• Shipped an internal push communication service with campaign targeting and delivery analytics, letting teams move off CleverTap and cut third-party tooling costs.
• Built an MCP-compatible internal dev server that gives LLM agents our UI libraries, coding guidelines, debugging context and code-navigation tools.

Associate Software Developer

• Built Android SDK integrations for glucometers and smart weighing scales, tuning BLE throughput, reconnection reliability, battery use and low-latency parsing of biometric data.
• Developed a gamified fitness experience on Google Fit and Apple Health that lifted user activity by 60% across trek-themed health journeys.
• Built platform tooling for localization, performance logging and storage migration; instrumented 4M+ user journeys with Sentry and moved AsyncStorage to MMKV for 30x faster reads and writes.
• Raised unit test coverage from 20% to 80% with Jest.

• Three accepted patches to React Native core improving TypeScript API correctness, accessibility role coverage, Modal native ref typing and Image srcSet parsing, each with regression or type tests.

An observability and debugging platform for AI agents that captures LLM calls, tool use, decisions and file edits, with step-by-step tracing, fork-from-any-step replay, visual diffs, cost visibility and eval suites that gate CI.

A fully local CLI and TUI for analyzing AI coding tool token usage across providers: heatmaps, cache ROI, model efficiency, cost-waste patterns, exports and shareable reports, with no telemetry uploaded.

A coding agent harness: one static Go binary with four default tools, a system prompt under 1k tokens, 13 model providers behind one API, and extensions that run as separate processes.

BTech in Computer Science, GPA 8.65