Project Background
Meta's Facebook Reels ranking team is preparing a model improvement launch intended to increase short-form video watch time before the end of the quarter. You are the senior Research Scientist leading execution across research, engineering, and product, and you have been asked to use this project to level up a junior Research Scientist who joined 4 months ago and has not yet led a launch workstream.
The team has 9 people: 3 Research Scientists, 3 ML Engineers, 1 Data Scientist, 1 Product Manager, and 1 TPM. The launch is urgent because leadership wants results included in QBR reporting, and the model change is expected to impact a core Meta surface with high visibility.
Key Stakeholders
- Reels Product Director wants an on-time launch with measurable watch-time lift
- Engineering Manager wants low operational risk and no increase in inference latency
- Junior Research Scientist wants ownership of a meaningful scope but still needs guidance
- Responsible AI reviewer wants fairness checks completed before ramp
- Experimentation infrastructure team has limited support capacity this month
Constraints
- Timeline: 8 weeks to launch recommendation
- Budget: no additional headcount; only $40,000 for offline labeling and analysis support
- Junior scientist can spend at most 60% of time on this project due to onboarding commitments
- One ML Engineer is shared 50% with Instagram Explore
- Any launch on Facebook Reels must stay within a 10 ms p95 latency increase and a 2% infra cost ceiling
Complications
- The junior scientist is strong technically but struggles to communicate trade-offs to non-research stakeholders.
- A dependency on the experimentation platform team may slip by 1 week because they are supporting a Threads launch.
- Early offline results show lift on watch time but a small decline in content diversity, creating a launch decision trade-off.
Your Task
- Build an 8-week execution plan that both delivers a launch recommendation and deliberately develops the junior scientist.
- Define what work you would delegate versus retain yourself, and why.
- Create a stakeholder communication and review cadence across research, engineering, product, and Responsible AI.
- Specify launch success criteria, rollback conditions, and how you would handle the diversity trade-off.
- Identify the top risks to both delivery and mentorship outcomes, with mitigations.
