Shreya Chinthala

• Built AI-driven backend services using Python, FastAPI, and vector databases to improve restaurant/menu search ranking and personalization, reducing query latency by 40% during peak dinner hours. • Developed scalable ingestion pipelines with Pinecone, LangChain, and async processing to embed and index restaurant menus, courier profiles, and marketplace signals, ensuring consistent retrieval accuracy across millions of listings. • Optimized caching and model-feature storage layers using Redis and PostgreSQL, enabling faster ETA predictions, courier recommendations, and order-fulfillment decisions while eliminating recurring timeout issues. • Deployed LLM and ML model updates through Docker and Kubernetes, powering 100K+ daily inference requests for real-time routing, demand forecasting, and support automation, with zero-downtime rollouts. • Monitored latency patterns and throughput using Prometheus and Grafana, identifying bottlenecks affecting delivery-time estimation, batching logic, and courier assignment, and implementing fixes that improved service reliability across regions. • Implemented OpenTelemetry-based distributed tracing across logistics and search microservices, improving root-cause analysis for routing delays, caching inconsistencies, and search-ranking regressions—cutting diagnosis time by 35%.

• Built REST API features using Java and Spring Boot, improving request handling logic and reducing repeated processing overhead that frequently slowed internal engineering tools. • Updated CI/CD workflows with Docker and GitHub Actions, shortening deployment cycles by nearly 30% and reducing build inconsistencies during weekly regression and integration processes. • Enhanced internal utilities using TypeScript and Node.js, improving data-processing reliability and enabling smoother coordination for teams depending on shared services for development activities. • Analyzed Datadog logs to isolate a recurring schema issue, applying fixes that lowered error occurrences by 18% during high-volume testing sessions.

• Developed supervised learning models using TensorFlow and PyTorch, improving classification accuracy by nearly 20% after refining feature inputs and adjusting training parameters across multiple iterative experiments. • Built feature pipelines with Python, Pandas, and NumPy, reducing manual preprocessing time by 45% and enabling faster experimentation cycles for continuing validation studies within the analytics team. • Deployed inference workloads using Kubernetes and Docker containers, ensuring predictable scaling during peak traffic and maintaining application response times within acceptable ranges for dependent downstream systems. • Monitored model drift through MLflow dashboards to schedule retraining windows effectively, helping maintain stable prediction quality and consistent performance across all deployed production environments. • Systematized recurring ingestion tasks using Airflow, lowering operational effort and improving consistency across data pipelines previously affected by manual scheduling issues and intermittent processing interruptions.

• Built backend APIs using Python, Node.js, and Express to streamline property-data operations, improving application responsiveness by nearly 28% across frequently accessed customer screens. • Integrated OAuth2 and JWT authentication flows, strengthening session reliability and reducing previously reported user-access disruptions by 40% during heavier activity periods on the platform. • Optimized SQL and NoSQL queries in PostgreSQL and MongoDB to improve data retrieval speed noticeably and support more stable reporting performance for internal analytical teams. • Containerized backend components using Docker to simplify environment setup, reducing onboarding complexity and improving deployment consistency across multiple development and testing environments.