TechiSeek

VendorsFor Vendors

Meta Andromeda: Powering the Next Generation of Ads Retrieval Through Model-System Co-Design

Meta Engineering

Meta Andromeda: Powering the Next Generation of Ads Retrieval Through Model-System Co-Design Blogs

Meta Andromeda is Meta's personalized ads retrieval engine, the first stage of the ads recommendation pipeline that narrows a vast pool of eligible ads to a small set of strong candidates in milliseconds before a more expensive ranking stage scores them. The operator-facing answer: change the retrieval model architecture and the system/hardware execution model together, not in isolation.

Examples

  • Since 2025, Meta Andromeda has delivered a 2.4% increase in conversions on Facebook.

Known limitations

  • Shipping ads between local CPU hosts and remote hardware accelerators saturates the network link.
  • The deep modeling quality of the ranking stage has historically been too costly to apply at retrieval's scale.
Visit Site

Recommended

Meta Networking at the 2026 OCP Global Summit: Open Innovation for Scale-Up, Scale-Out, and…Meta Networking at the 2026 OCP Global Summit: Open Innovation for Scale-Up, Scale-Out, and…Meta Engineering Python Typing Survey 2026: AI Tools and the Shifting Checker LandscapePython Typing Survey 2026: AI Tools and the Shifting Checker LandscapeMeta Engineering Catching Bugs Before They Land: JiTTesting’s Human-on-the-Loop Approach at Full ScaleCatching Bugs Before They Land: JiTTesting’s Human-on-the-Loop Approach at Full ScaleMeta Engineering Trained on 100,000+ Voices: Deepgram Unveils Next-Gen Speaker Diarization and Language DetectionTrained on 100,000+ Voices: Deepgram Unveils Next-Gen Speaker Diarization and Language DetectionDeepgram System UIconsSystem UIconsCss Tricks Building a Remote MCP Server: OAuth, Tool Design & Lessons from 4,000+ AI Queries | MotherDuckBuilding a Remote MCP Server: OAuth, Tool Design & Lessons from 4,000+ AI Queries | MotherDuckMotherduck Trigger automation - MintlifyTrigger automation - MintlifyMintlify Docs-as-code solutions for API teams: how to choose the right platform in 2026Docs-as-code solutions for API teams: how to choose the right platform in 2026Mintlify Mintlify Alternatives: What to Consider (and Why There's No True Substitute)Mintlify Alternatives: What to Consider (and Why There's No True Substitute)Mintlify Q&A with Deepgram’s New CPO, Ed AnuffQ&A with Deepgram’s New CPO, Ed AnuffDeepgram New Spanish and Turkish Language Models and Updated General Models - Deepgram Blog ⚡️New Spanish and Turkish Language Models and Updated General Models - Deepgram Blog ⚡️Deepgram Redis Achieves ISO 27001 Cybersecurity CertificationRedis Achieves ISO 27001 Cybersecurity CertificationRedis Intel Unveils Next Generation Neuromorphic Computing ChipIntel Unveils Next Generation Neuromorphic Computing ChipThenewstack Three Key Design Considerations for an Event-Driven ArchitectureThree Key Design Considerations for an Event-Driven ArchitectureThenewstack Connect Any Git or Mercurial Repo to Pulumi with Custom VCSConnect Any Git or Mercurial Repo to Pulumi with Custom VCSPulumi Neo Integrations: MCP Servers and Cloud CLIsNeo Integrations: MCP Servers and Cloud CLIsPulumi The Go Memory Model - The Go Programming LanguageThe Go Memory Model - The Go Programming LanguageGo What is a CUDA Thread Block?What is a CUDA Thread Block?Modal Anthropic integration with Modal brings scalable compute to Claude ScienceAnthropic integration with Modal brings scalable compute to Claude ScienceModal Building an RL theorem-proving workflow on ModalBuilding an RL theorem-proving workflow on ModalModal