Co-founder of Skyvern. YC S23. Skyvern helps companies automate workflows in the browser with AI.

Built a ML Platform for marketplaces to help them improve their revenue through better product ranking. Went through YC with this idea, drove the company to $50K ARR + $240K in sales commitments, and pivoted away from this product

Built a tool to help onboard software engineers. Pivoted.

Founded the Search Engineering team at Gopuff. Our primary responsibility when I joined was to move off a 3rd party Search system and bring it in-house. We had a rudimentary Search system build using text-matching in Elasticsearch and a semantic ranking model capturing relevance and product quality. We supercharged this system with some sophisticated query understanding with BERT models, real-time ranking XGBoost models layered with business logic to overtake the vendor we were working with.We also built a new recommendations system capable of serving real-time personalized recommendations to users in <100ms.In addition to this, we expanded this system to be used across all product ranking and recommendation surfaces across Gopuff.The impact of this project is nearly 2 Million dollars in bottom line savings for the company, and >$20 million dollars of top line incremental revenue impact, as measured by A/B tests.In addition to my technical performance, I also manage an Engineering team that continually strives towards pushing the limits of Search and Gopuff.1. Managed a team of 5 people and created foundational processes that weigh collaboration and productivity2. Built a new ad-hoc requests process to help the team operate more efficiently with fewer distractions3. Set up a new design review process in the team to facilitate farming for dissent and ensure engineering excellence4. Led the team to deliver a few key foundational features such as Search Filters, Product Substitutions, etc.5. Built out new processes that weigh product needs (tight feature estimation, dates on product delivery) and engineering needs (developer productivity, developer creativity and satisfaction, and growth)6. Proposed, built, and delivered high quality ML Infrastructure being used across several teams at Gopuff7. Leveraged the above infrastructure to deliver $20M+ of incremental revenue to the business through iterations on Relevance and Ranking projects

Founded the Machine Learning Infrastructure team, with a focus on creating a robust Machine learning platform to be used across all departments at Faire. The primary focus of this team was to solve problems for search, and "platformize" them so the solution could be used by other teams.Examples of the areas of ownership for my team were:1. The online and offline feature store 2. The model training pipeline, largely focusing on XGBoost models for low-latency predictions3. Model monitoring systems including anomaly reports, prediction drift detection, real-time feature drift4. A Real-time model inference system capable of making thousands of predictions with tens of thousands of features in <100ms, with single model inferences being measured as low as single-digit millisecond performance5. Overall system health metrics (Online -> Latency, Accuracy, Memory, Real-time features, etc), (Offline -> MRR / NDCG over time, Search conversion over time, etc)Areas of my impact included:1. Founding the team from ground up2. Building out a career ladder for Machine Learning Engineers3. Defining the job description for the role, including day-to-day responsibilities4. Founding an Engineering-driven roadmap spanning several years of planned work to ensure the systems are able to scale with the company's hypergrowth, especially as the company transitioned from "moving fast and breaking things" to "moving fast, with quality in mind to serve our customers"5. Managing and mentoring a small team to empower them to reach their true potential

Designed and built an online Feature Store for Faire, capable ofretrieving tens of thousands of features for a particular user requestin under 50ms1. Architected Faire’s real-time ranking system that includes featureretrieval, online feature computation, model evaluation, and featurerecording. The system scores hundreds of products and serves recommendations in under 100ms2. Implemented an end-to-end monitoring and alerting framework forthe real-time ranking system that allowed us to monitor system health and guarantee uptime3. Optimized the online feature store with state-of-the-art serializationand compression scheme, reducing latency, storage, and memory usage by over 50%4. Built the foundational “real-real time” feature generation systemthat tracked user session behaviour as they browsed our site tosupercharged our machine learning models5. Deployed a sophisticated multi-armed bandit scheme aimed toimprove the experience of new brands on Faire. Improved newbrand conversion by over 20%, a key company initiative at the time6. Trained a brand new search real-time ranking model that improvedthe top line revenue of the company by 2%, translating to 10s of millions of revenue impactRead more about the system I designed and built here: https://craft.faire.com/building-faires-new-marketplace-ranking-infrastructure-a53bf938aba0https://craft.faire.com/real-time-ranking-at-faire-part-2-the-feature-store-3f1013d3fe5d

Working with Android and security algorithms to preserve user privacy.

Worked in J2EE to develop various features for core Drivesync platformWrote code in JSP, JavaScript, and CSS for front end changes Utilized Spring technologies for dependency injectionWorked with Hibernate and HQL to manage Entity beans and DAOsMaintained maven integration to ensure all of the unit tests were running properlyFollowed the scrum methodology with weekly sprints Wrote unit tests in JUnit and JMock to ensure proper test coveragePerformed some regression tests to ensure each part of the software was up to standardTechnologies used: Java/J2EE, Ant, Maven, SQL, JSP, CSS, JavaScript, Hibernate, HQL, jBoss, Spring, Eclipse, Jira, Hudson, Jenkins, Subversion, Scrum, JUnit, JMock, PostgreSQL, MySQL, OpenLDAP.

Debugged various problems using WireShark to perform packet analysis on various message types including SIP Messages and TCP/IP Messages Worked in JUnit and SIP Unit to automate Instant Messaging and SIP VoIP calls Performed regular sanity testing and regression testing to ensure various features in IMS, Standalone, and SSL infrastructures were up to standardMentored a new full-time employee and familiarized them with the environmentTechnologies Used: Java, SIP, TCP/IP, RTP, TLS, UDP, WireShark, Red Hat variant of Linux, JUnit, 3GPP IMS Server Configuration(I, P, S-CSCF, HSS, A2 Server), SSL Infrastructure, SIP Unit.

Developed in Java to add features and fix defects to their primary productWorked in a Client-Server architecture; implemented changes both server side and client sideMaintained the Hudson/Jenkins build server by ensuring that the ANT build scripts were functionalManaged the Scrum team’s weekly point capacity (weekly team quota) Technologies Used: Java, Oracle, Ant, Hudson/Jenkins, Guice, Jira, HP Quality Centre, BugZilla, EasyMock, PowerMock, JUnit.