“During my internship, I have done work in programming, design, product management, all of which are at the core of user experience.”
Overview
I worked on the recommendation team, which design and implement Hulu’s personalization system. My initial focus was to improve and implement various Scrum methods ranging from Daily Scrum to Scrum Review in order to complete the feedback loop and to increase team efficiency. Through new findings in user research, my focus has shifted to enhancing user experience through algorithms, interface design, and many other ways. Towards to the end of my internship, I have also led a small team to finish the tech-blog rebranding task.
During my internship, I had the chance to work with our recommendation team and needed to gather resources and information from search team, data-science team and Dev-Ops in order to ensure the projects can finish smoothly
Projects
Recommendation Team Scrum Implementation
Tech Blog Re-skin
Recommendation Infrastructure
Softwares
Sketch Jira Adobe Suite Python
What I Did
Project Management, UI/UX Design, Crawling
Scrum – Project Goal
- Engage the recommendation team with the concept of Scrum
- Increase the internal transparency
- Implement Scrum Review, Scrum Planning, and daily Scrum teamwide
As an acting team manager, I regularly reported the output of Scrum stakeholders outside of the team, increasing our exposure and the chances to collaborate with other teams. To ameliorate the potential unwillingness of implementing Scrum, I did not use the term “Scrum” in many activities to get people onboard.
To execute successfully and smoothly, I add a new method(Daily Scrum, Scrum Planning and Scrum Review) every two weeks. In this way, the feelings and the effort of the Scrum can be preserved. By the end of my internship, the Scrum Review has more than 10 projects to show with more than 60 attendance.
As a Scrum Master, I had to keep track of the team scrum tickets and to be aware of the potential conflicts. I also had to understand team dependencies in order to hold meetings. Team members, on the other hand, needed to fill tickets and attend necessary Scrum meetings.
By filling a completed ticket, a task description in Scrum system, team members can check, track the progress of anyone’s work inside a team, building a sense of community
Scrum – Project Accomplishment
- Team velocity has increased by 20 percent
- Outside stakeholders are more likely to see and check the status of the project
- Project awareness of the team has also escalated
Scrum Software
Fast Tracking
Putting all the tasks in one place and greatly reduce the potential risk of duplicates.The ticket response time(from being created to be resolved) has also reduced significantly.
property of Atlassian
Improved Efficiency
Build Awareness of the Team
With such method being implemented, all team members started to wrote more documentation for their own reference, resulting in a postivie feedback loop.
20
More Tasks Being Resolved
400
Tickets With More Complete Info
32
Team Velocity Increase
Recomendation Infrastructure – Project Goal
- Increase user engagement
- Build a new entity mapping information among movies, shows, actors and so on
- Solve existing problems
My regular involvement with the recommendation team concerns about building the world-of-class recommendation system using machine learning. I have participated as a product manager to offer algorithm design suggestion and code review
To build a new recommendation system, we adopted the use of CF-NADE, a context-aware classifier that connects the algorithm suggestion with the environment that the user surrounds. The result is a low RMSE of below 80 percent.
I have also engaged with the cold-start problem where typically happen when there is not enough data to populate user’s home studio.
I have participated in offering code suggestion to algorithms and involved in code-review process to contribute my engineering skills. For example, the initial recommendation screen of any user would be implemented by a (Depth-first-search)DFS method where shows with high similarity inside one category would occupy the whole screen. However, such method has obvious disadvantages where people do not have the chance to experience other shows. Therefore, I offered a suggestion to switch to a Breadth-first search(BFS) to combine with the existing algorithm.
Due to NDA, I can not talk about the detail of our infrastrucutre.
Reco Infrastructure – Project Accomplishment
- All Tasks were on track
- Every tasks were finished on time with adequate external resources
- User engagment has increased from our monitor results
It was wonderful to witness the birth of CF-NADE through our work at Hulu to benefit millions of users worldwide.
17
Increase in Subscription Members
173
Brand Love Index
3
More episodes seen by each per month