Sunnyvale, California, United States
Hi there! My name is Emil, and I am a software engineer at Meta. I work on Facebook Search Ranking and have grown an expertise over location-based searches. While I am officially on Ranking, I’ve worked closely with sister NLP and Retrieval teams to deliver solutions with real impact on the Facebook Search experience for location-based queries and have gained deep knowledge on how end-to-end search systems operate. I’ve also dug deep into general feature engineering for our models. I enjoy working with deep complicated code stacks and unraveling how large scale systems work and deliver their experiences efficiently and configurably. Any chance to be a code detective and diagnose potential pitfalls and improvement areas, sign me up! Before working at Meta, I was a software developer intern at Epic Systems. I worked on revamping their MyChart ToDo UI, working closely with UX researchers and customers to understand customer requirements and deliver a delightful experience. I loved working on a product that was health-focused and feeling like my work would help the people I see in my day-to-day life.
- Drove quarter-long Location Relevance sprint: investigated key problem areas in end-to-end search stack, created investigative documents summarizing these findings into buckets (NLP signal failure, misconfigured retrieval generators, location-unaware ranking models), and led meetings to consider all options and align on one solution to execute on. Reduced third-party rater-based location failure rate from 14% to 5.8% (58% reduction) - Utilized an experimental “near me” signal from the Query Understanding team and existing document-side location signals to develop a new retrieval filter pattern to improve the “near me” query experience. Reduced global search reformulations by 0.69%; increased global search engagement rate by 0.63%; and improved precision/recall metrics on “near me” production-sampled query test set by 35.3%/24.4% - Developed LLM-based location search metrics that rate the search experience similar to the topline relevance raters; iteratively prompt-engineered to ensure the results are comparable to raters; configured parallelization to fine tune the prompt rate to avoid rate limiting but still be performant; saved engineers up to 4 hours of testing time when validating location-relevance improvements with their changes - Authored documentation and workflow diagrams of how location search works from NLP to Late Stage Ranking - Created onboarding bootcamp tasks for new team members that explain how to debug across the search stack
MyChart To-Do UI Redesign: Enhancing the UI of a list of tasks that aim towards addressing a health goal or problem - Identified key problems of low user engagement with overdue tasks caused by excess navigational options and lack of application context caused by missing care plan labels - Inquired with direct, actionable questions in project check-in meetings for feedback on the current design and quickly iterated according to the responses to keep up a constant stream of exploration and development - Led a customer call with Ochsner that revealed how to best move the content of the overdue tasks to simplify navigation and implemented this solution such that it was maintainable despite the change in workflow - Analyzed the existing React structure of the project to create a generalizable care plan header that can be easily utilized throughout the codebase to provide context to different tasks regardless of the view settings a user is in - Developed multiple views that solve the key problems identified but allow users to decide whether they want to see their tasks primarily by the time they are due or the care plan they are in, increasing user customizability - Experimented rapidly on an animated weekly calendar that would allow the user to see their tasks by a certain day, gamify the process with progress rings to encourage user engagement, and quickly express which days have overdue tasks to encourage overdue task completion