Developed and optimized PayPal’s Omni-Channel Platform by implementing strategic caching for LLM prompts and responses, significantly reducing redundant API calls. Led the architectural migration of customer summarization models from a separate Python-based Azure GPT-4o-mini service to a native Java implementation using the Gemini-2.5-flash Gen-AI SDK. Used LLM API's to summarize customer-data, enabling seamless cross-channel payment experiences. Collaborated with cross-functional teams to design and deliver scalable, reliable backend solutions, optimizing integration and transaction flows across diverse retail environments. Additionally, engineered Agentic AI conversational workflows utilizing Google Dialogflow (GDF) to unify cross-platform customer chat flows.

Developed and implemented an automation and analysis tool for the lab, runs on a cloud server, in a colocation facility – an Assembly Line - which monitors lab tool data. For this Data Monitoring system I was responsible for designing a PostgreSQL-based data monitoring system and integrating Kafka for real-time metrics tracking (e.g., throughput, error rates, and status).

Extracted data was cleaned by removing and replacing the missing values, removing outliers and invalid entries. Standardizing the tool names, summarizing data into hourly aggregates, calculating derived metrics like uptime_percentage or error_trend (rolling 24-hour error rate average), resampling data to fixed intervals (5-minute buckets) for dashboard compatibility, and flagging anomalies using thresholds (error_rate → status = "needs_inspection") by integrating with Kafka to trigger alerts in real time. Write this processed data to PostgreSQL for dashboards for analytics.

Built RESTful APIs for data visualization and implemented robust database models, enhancing retrieval speed by 30%. This tool improved the labs scalability and performance by 23%.

Working on addressing the "cold-start" problem in containerized applications by meticulously analyzing latency issues in Kubernetes clusters and recently achieved significantly improved performance by optimizing container preload, caching, and eviction based on the machine learning application's request patterns.

Computer Science Engineering with Data Science Specialization.

Academic Project: Implemented a system to detect bot requests coming on an auctioning site. Preprocessed, trained 260MB dataset of requests, and tested over 10 different machine-learning models to get a 96% accurate prediction. Also simulated an auctioning site that was bombarded with bot requests to visualize the detected bots on the bidding page. Tech stack used: Python, ML Model Libraries(scikit-learn), Django, HTML, CSS, RESTful API, MongoDB, Postgres DB.

Internship. Implemented a critical feature aimed at optimizing manufacturing assembly line operations by identifying and addressing bottlenecks. Leveraging the assembly station and production details dataset, I designed a system that strategically aggregated data pertaining to processing times. This data served as the cornerstone for pinpointing any production bottlenecks, thereby enhancing overall system efficiency.

The aggregated dataset, a pivotal component of our solution, was meticulously crafted and stored on the Databricks cloud platform. Given the high-paced nature of our assembly line, where approximately 20 million records were generated within a mere 24-hour timeframe, this dataset aggregation was a formidable task. It was essential for us to run this batch processing operation every 24 hours to ensure we collected and processed the latest assembly data.

In terms of technical implementation, I spearheaded the development of a highly scalable ETL (Extract, Transform, Load) pipeline. This pipeline, crafted using a combination of Scala, Spark, and Databricks technologies, not only processed the data efficiently but also exhibited an exceptional level of robustness and scalability. Furthermore, to ensure the reliability of our codebase, I maintained a rigorous approach to testing, achieving a remarkable 100% code coverage.

This initiative bore testament to our commitment to continuously enhance the assembly line's performance. By systematically identifying and addressing bottlenecks using cutting-edge technology and data-driven insights, we made significant strides in streamlining the manufacturing process, ultimately contributing to a notable improvement in the company's assembly line operations.