Meta's Feed Ranking & Recommendations team is seeking a senior Technical Program Manager to lead one of the most technically ambitious programs in our recommendations ecosystem: unifying two historically separate large-scale ranking stacks — content-first and interest-first recommendation — into a single, end-to-end ranking system. Today these stacks maintain parallel models, pipelines, and serving paths; the future state is one common foundation spanning retrieval, model serving, value modeling, and the control layer, built on a modern recommendation-as-a-service platform.

This is a rare "re-architecture in flight" program: you will drive the consolidation of production ranking systems that serve billions of people, while the systems continue to run and while topline results must hold. The work sits at the intersection of large-scale machine learning, serving infrastructure, and product impact, and requires a TPM who can go deep technically, operate in a highly ambiguous problem space, and align many engineering, infrastructure, and product teams around a shared multi-quarter roadmap.

The impact you'll make:

• SOTA Models: collapse duplicated models, pipelines, and tooling into one stack, materially reducing the cost and time to ship ranking improvements.
• Capacity savings: consolidate serving and training footprints to reclaim significant GPU inference (and downstream training and dataset) capacity — a first-order priority in a capacity-constrained environment.
• Faster, better topline: establish a single, modern lever for ranking innovation so future model and product wins land faster and with higher quality.

Work complexity — what makes this role hard

• Migrating live, high-scale systems: you are re-architecting production ranking systems without downtime and without regressing metrics that leadership watches closely.
• Competing hard constraints: latency, capacity, launch neutrality, and delivery timelines are frequently in tension; you will broker principled tradeoffs across them.
• Deep technical ambiguity: the target architecture is still being defined in parts; you must frame ambiguous problems into a concrete, sequenced plan and adjust as the technical picture evolves.
• Broad, matrixed collaboration: success depends on aligning many ML, infrastructure, data, and product teams — none of which you own — around one roadmap and one set of priorities.