Headline

Voice AI & Distributed Systems | Ex-Amazon | Ex-Toyota Connected | IIT Kharagpur

About

Backend Software Engineer and IIT Kharagpur graduate with experience at Toyota Connected and Amazon building scalable distributed systems, cloud-native services, and intelligent automation platforms. I’ve designed large-scale production backends in Java, Go & Rust, architected event-driven systems and serverless workflows with AWS Lambda, DynamoDB, and SQS, and delivered impactful features including GenAI-driven automation, voice-controlled features for the Next-Gen Lexus virtual assistant, and real-time weather alerting systems. My work bridges backend engineering, cloud infrastructure, and applied machine learning, with a focus on creating reliable, high-performance systems that scale across platforms.

Location

Bengaluru

AI stack

claude code

claude

codex

cursor

gemini

claudecode

Work experience

Founding Engineer at hiRobin, Mar 2026 - Present · 7 mo, Bengaluru, India, • Real-Time AI Voice Platform: Architected and built the end-to-end speech-to-speech AI platform pow- ering 500K+ calls/day across 100K+ daily active users, delivering <2s p95 conversational latency while serving 36M+ production voice calls on multi-region AWS infrastructure. • Real-Time Voice Infrastructure: Engineered voice infrastructure spanning Gemini Live and cost-optimized ASR + LLM + TTS pipelines, implementing bidirectional audio streaming, intelligent prewarming, session migration, and resilient WebSocket recovery for production-scale AI conversations. • Agent Orchestration Engine: Architected and built the orchestration engine powering autonomous AI phone agents with tool-calling, task scheduling, retries, batching, and callback recovery, executing 700K+ outbound calls across 300K+ autonomous task workflows. • LLM Evaluation Infrastructure: Built the complete AI evaluation stack from scratch, including grounded regression suites, synthetic AI callers, production quality scoring, instruction-adherence analysis, and Voice Quality dashboards evaluating 32M+ production conversations. • Distributed Backend Systems: Scaled core services using FastAPI, PostgreSQL, MongoDB, and Redis, eliminating distributed race conditions, optimizing cross-region request paths, and implementing cluster-safe coordination for real-time workloads. • AI Cost Optimization: Reduced Gemini Live inference costs by ∼42% through prompt redesign, input- audio gating, canned responses, and pipeline optimization while maintaining 0.87/1.0 internal quality. • Experimentation Platform: Designed and shipped an in-house A/B experimentation framework with deterministic sticky assignment, live configuration, zero-storage variant allocation, exposure telemetry, and production-safe rollouts across backend services. • Production Observability: Established core backend infrastructure with deployment pipelines, structured observability, distributed request tracing, resilient logging, and production database workflows, using call-level correlation and PII-masked logs to improve reliability and debugging across services.

Software Development Engineer at Toyota Connected India, Mar 2023 - Mar 2026 · 3 yrs 1 mo, Bengaluru, Karnataka, India, • Large Action Models: Architected and implemented a device-agnostic backend enabling LLM inference for Generative AI–driven UI automation with function calling; delivered a 0→1 Android automation system with <6s per action (high-performance for UI automation) using Amazon Bedrock & on-device LLM reasoning. • Scalable Automation Architecture: Built a fully containerized execution framework with Docker for re- producible builds and efficient deployment; extending the LAM framework to integrate with Toyota One App(Android & iOS) and the Lexus Virtual Assistant to enable LLM-powered UI interaction flows. • Lexus Next-Gen Virtual Assistant: Built high-performance, low-latency (<2s) voice features using Cap’n Proto, delivering fast and accurate interactions across HVAC, Weather, Settings, and Vehicle Info. • Severe Weather Alerts: Designed and deployed a real-time geo-targeted severe weather alerts system, using AWS SQS & Redis, reducing alert-delivery latency from 3s to <1s via smart caching. • Smart Climate: Engineered adaptive HVAC control logic for air conditioning, fan speed, and defrosters, optimizing zone-based microphone processing for precise, context-aware voice recognition.

Software Engineer at Amazon, Jun 2022 - Mar 2023 · 10 mos, Bengaluru, Karnataka, India, Java Microservices Architecture: Architected and built backend components in Java using dependency injection (Google Guice) enabling regional product-control workflows; authored and developed the auditing subsystem with AWS Lambda, DynamoDB Streams, and SQS to ensure reliable, real-time event tracking. Infrastructure as Code: Created, deployed and managed backend infrastructure with AWS Cloud Development Kit ensuring reproducible and CI/CD-aligned production environments. Operational Excellence: Implemented monitoring, alerting, and anomaly detection with AWS CloudWatch, driving proactive incident prevention and faster resolution cycles. Resilient Event Processing: Designed and implemented retrial mechanism for extended failed event retention from 24 hours to 14 days, improving audit reliability.

Education

Bachelor's degree, Indian Institute of Technology, Kharagpur