Property QnA AI Agent

Launched LLM-powered personalized itinerary recommendations (flights/lodging/activities); drove retention and conversion via internal and third-party data integration; scaled globally to 3 brands, 7 languages, 27 locales.

Designed backend systems (Kotlin/Spring Boot/MongoDB/Kafka) for scalability and availability.

Enhanced GenAI performance (accuracy/completeness/relevance/consistency) via prompt chaining, fine-tuning, model evaluation and seamless migration; reduced latency and ensured AI safety.

Leveraged MongoDB, asynchronous processing, and Kafka for efficient data flow and scalability.

Developed Spinnaker CI/CD pipelines; implemented Splunk/Datadog monitoring for proactive issue resolution.

Collaborated with cross-functionally (product/ML/design) to align project goals, drive innovation and achieve impactful results.

Productionized a two-stage model workflow (in-house RoBERTa + GenAI) to detect platform policy violations (Python/TorchServe/Docker/K8s/AWS).

Built real-time inference services and model packaging deployed via a robust RESTful architecture.

Marked the company’s first adoption of GenAI, enhancing travel planning experience through AI assistants.

Supported 10+ brands, 15k agents and 100M+ million customer interactions daily (React, Node.js).

Fine-tuned casual language models (DialoGPT) using in-house data sources to engage travelers with human-like dialogue.

Built production ETL pipelines for derived datasets used in house daily and contributed to open source repositories.

Predicted ad clicks using predictive models (Logistic Regression, Random Forest, XGBoost) via scikit-learn and identified key features to increase ad revenue. Final model achieved 0.89 AUC and 0.81 accuracy.

Delivered reports using R (Shiny, ggplot) to support multiple business decisions for the management teams.

Developed and deployed an end-to-end data analytics and visualization dashboard that provides insights into online sales of company products.

Presented app to company executives and used the Agile software development lifecycle to incorporate their suggestions.