Project Background

You are a Research Scientist at Meta who led a high-impact publication on a new multimodal modeling approach with strong offline gains on image-text understanding. Leadership now wants to convert the research into a production-ready capability for Meta AI and a downstream ranking use case in Instagram within 16 weeks, while preserving the novelty needed for conference presentation and follow-on publications.

The project team includes 6 research scientists, 4 research engineers, 2 product engineers, 1 TPM, and shared support from Responsible AI, Legal, and Infra. The work is urgent because a competing lab is rumored to be close to publishing similar results, and Meta leadership wants an internal demo before the next org review.

Key Stakeholders

Research leadership wants scientific rigor and publishable novelty. Product and engineering leaders want a scoped launch with measurable user impact. Responsible AI and Legal require additional reviews before any model touches user-facing surfaces. Infra wants to limit training and serving costs on shared GPU clusters.

Constraints

You have a 16-week timeline, a compute budget capped at 1.2M GPU-hours, and no additional headcount. The model must support inference latency under 180 ms p95 for the Instagram use case and cannot degrade existing ranking quality by more than 0.5% on key guardrails. A privacy review must be completed by Week 6, and production integration depends on one shared feature pipeline owned by another team.

Complications

1. The shared feature pipeline team is already committed to a Threads launch and can only support one integration window in Weeks 7-8.
2. Early red-teaming found elevated failure rates on low-resource languages, creating launch risk for global surfaces.
3. Your coauthor wants to hold back some implementation details until the paper is accepted, while product leadership wants enough transparency for maintainability.

Your Task

1. Build a 16-week execution plan from publication-stage research to limited production launch.
2. Explain the core trade-offs you would make across research ambition, launch scope, latency, and risk.
3. Define launch success criteria, guardrails, and a rollback plan.
4. Identify the top execution risks and how you would mitigate them.
5. Recommend how to align stakeholders when publication goals conflict with productization needs.