-
Software Engineer @ Copart
Aug 2024 — ongoing | Dallas, TX
Developed an in-app message notification service using Kafka, NChan, Redis, Cassandra, and Spring boot.
Scaled to 1M daily notifications by leveraging Kafka lag metrics for consumer autoscaling,
boosting the user engagement rate by 30%.
Implemented CDC-based data pipelines to sync data from Cassandra to Iceberg
and build an analytics dashboard in Tableau, leveraging queries executed by Trino query engine
to summarize daily notification trends and identify anomalies.
Integrated customer support ticketing into all user-facing apps with Spring Boot, Solr and Kafka,
enabling direct ticket creation and built a dashboard over the Solr collection for internal resolution
tracking, reducing customer grievance SLAs by 20%.
Migrated all existing Redis operations to leverage pipelining across all internal services,
reducing network utilization by 30% and Redis cluster’s CPU utilization by 40%,
while decreasing rate limiter’s average latency from 70ms to ~40ms.
Upgraded all member microservices from Spring 2.7.3 to 3.2.2,
ensuring compatibility with Java 21 and GraalVM JDK.
Technologies used:
-
Graduate Research Assistant
Jan 2024 — Aug 2024 | Richardson, TX
Developed a chatbot application for UTD-related queries using Retrieval-Augmented Generation (RAG),
with UTD documents and web page content stored in S3 and their vector representations managed in
OpenSearch for efficient query similarity search.
Deployed a fine-tuned Llama 3.1 model on AWS Bedrock, enabling accurate and context-aware responses
to user questions.
Implemented a FastAPI backend with LangChain to process chatbot queries,
scaling dynamically based on the traffic load using EKS.
Implemented data pipelines in Python and orchestrated them using Airflow, to automatically retrieve
new documents from the S3 bucket and periodically re-scrape UTD websites,
ensuring the OpenSearch index remained up to date with the latest content.
Configured S3 Lifecycle rules to automatically archive document and user chat data older
than 3 months into S3 Glacier, optimizing storage costs while maintaining data retention compliance.
Technologies used:
-
Software Dev Engineer 2 @ Swiggy
Oct 2022 — Jul 2023 | Bangalore, India
Developed a model feature store in the in-house data science platform to cache precomputed features in
Redis/DynamoDB, based on individual model’s latency and RPM needs, reducing response times
by 80% on average, and boosting revenue by $20M/month.
Designed PyTorch model serving capability within the Spring Boot codebase
using the Deep Java Library and Nvidia Triton for our in-house data science platform,
enabling PyTorch integration for production serving and reduced TensorFlow model memory usage by 30%.
Implemented robust platform cost monitoring mechanisms,
leading to a significant reduction in the Data Science team's compute usage by approximately 45%.
Developed an observability tool using Databricks, Kafka, and Spark, hosting and monitoring 80+ deployed
data science models in real-time, improving governance and reducing response times by 20%.
Technologies used:
-
Machine Learning Engineer 1 @ Swiggy
Dec 2021 — Sept 2022 | Bangalore, India
Expanded the reach of smart push notifications to 20 million users, while drastically reducing Spark
job runtime from 9 hours to 2 hours and cutting compute costs by ~75% (from $688.74 to $166.84).
This resulted in a daily order increment of 117.6 on average and a 14% decrease in push notification uninstall rates.
Improved the spell correction TensorFlow model inference time to under 100ms,
successfully tested at a load of 1000 requests per second.
Developed a proof-of-concept (POC) for an insights dashboard
aimed at providing actionable competitor insights to restaurant partners.
Technologies used:
-
Data Science Engineer @ Stylumia
Jun 2020 — Dec 2021 | Bangalore, India
Led development using SQS, Elasticsearch, HuggingFace, and FastAPI, achieving 1200 RPS throughput
and 30 ms P99, which increased user engagement by 15%.
Automated the image-tagging process for customer-specific report generation,
cutting lead time by 50% and resulting in an 18% increase in customer conversion rate.
Spearheaded the use of Jenkins for microservice deployment automation through CI/CD pipelines
and established a culture of thorough code reviews, enhancing deployment speed and code quality.
Technologies used:
-
Research Intern @ Siemens
Jan 2020 — Jun 2020 | Bangalore, India
Engineered an attention-based algorithm to estimate weight and water content from potato images,
increasing estimation accuracy by 8% and reducing processing time by 30%.
Integrated the algorithm into an internal food quality assessment tool being prototyped at Siemens,
using Django and Docker for seamless operation.
Technologies used:
-
Intern @ Flutura
Jun 2019 — Jul 2019 | Bangalore, India
Contributed to the development of the chatbot feature on Flutura's Cerebra platform. The project used
the Django REST framework, PostgreSQL, and DialogFlow for the backend, and Angular for the frontend.
Responsible for adding new features and managing existing dialogue options in the backend part of the project.
Technologies used: