Romil Shah Email & Phone Number

Sunnyvale, Ca, Us

Menlo Park, Ca, Us

## Meta AI AgentsBuilding AI agents framework for digital twins and custom agents https://about.fb.com/news/2023/09/introducing-ai-powered-assistants-characters-and-creative-tools/## Reels RecommendationsEngineering lead for the Video Recommendations AI team with the purpose of maximizing user, creator and platform value. + Lead team of ML engineers and datascientists to drive significant year over year improvements to user experience and creator success on Facebook Video ranking on News Feed and Watch. + Build state of art deep learning modeling techniques for capturing niche Interests, Popularity and unbiased Ranking+ Design large scale ML infrastructure to power Video Recommendations for 1.25B users

Mountain View, California, Us

Architect of the 3 search verticals at Bloomreach: Semantic Search, Product/Content Ranking andPersonalization+ Models built by team drove revenue lift of average 5%, beating industry incumbents and major competitors in numerous statistical algo evaluations.+ Led the Human+AI effort, by driving interpretability and influenceability in ML powered applications.Scaled the team from 2, to a geographically distributed team of 10 engineers. Semantic Search+ Worked on transitioning search from a text-based match to a more intelligent matching strategy by using Semantic Understanding of products, content and user queriesPersonalization+ Worked on building personalization signal for search as well as algorithm to power personalized widgets.+ Designed and productionized 3 versions of Personalized Search and Recommendations algorithms.Product and Content Ranking+ Worked on building a machine learnt ranking platform which enables internal data scientists as well as customers to customize ranking optimized for search revenue goals.+ Co-Authored and Productionized all versions of search ranking algorithm. Built a Learning-To-Rank system that adapts across customer verticals

Mountain View, California, Us

Responsible for developing scalable algorithms and systems for supporting the search & search analytics product.Search Quality- Built the first version of scalable search recall and ranking with user performance feedback.- Designed the ranking and recall framework which is used today as the platform for all ML ranking models.Search Analytics- Built the AB testing infrastructure for algorithm evolution, established testing best practices, trained multiple engineers and data analysts on ab testing best practices- Implemented system to perform clickstream analysis of millions of user events daily. Analyzed data then used to evaluate usage of search and search features.- Author of core finding methods algorithms which are now foundational to Bloomreach internal and external search attribution.

Mountain View, California, Us

Redmond, Washington, Us

Part of 2 member team that owns Windows Shell components and works on feature additions and OS security . Windows Shell Development Team  Worked on Change Requests across user and kernel space, in the areas of Windows Explorer, Windows Search, Profile service and DirectWrite.  Redesigned User Environment and Windows Search components for Windows 7 and downlevel.Microsoft Security – Shell Team  Analyzed and Fixed several security vulnerabilities for Shell components, which could lead to security attacks on Windows machines.  Worked on publicly known security vulnerabilities like Baby Picture and DirectWrite font vulnerabilities.

Noida, Up, In

Developed high performance interception systems on UNIX platform, for intercepting voice and data connections over GSM network.GSM Mass Monitoring System Contributed to developing a Distributed telecom monitoring system which captures raw data from large number of sites, processes and co-relates the data to provide near real-time call and data interception. Designed and Implemented Call Engine and Data compression modules.Challenges: Designing a data analysis framework that could handle real-time data of 12000 calls per node. Designing modules which are re-entrant and highly optimized for performance, scalability and multi-threading.