AI Infrastructure at Platform Scale: Powering Fast, Trusted and Compliant Personalisation

In every digital category, from media to marketplaces to professional networks, the competitive edge now lives in the infrastructure that power Machine Learning use cases like ranking, relevance, personalization, etc. Experiences must feel instantly relevant, stay reliable under bursty demand and meet tightening governance and safety bars across regions.

Against that backdrop, Sayantan Ghosh, a Senior Engineering Manager at LinkedIn and the co-inventor of a widely-cited patent, “Correction of user input,” leads AI/ML, Data and Systems teams that underpin feed ranking, personalization, discovery and high-availability delivery. His remit: modernize infrastructure, embed safety alongside scale and translate ML capability into durable product outcomes across creator distribution and member engagement.

Rising AI Adoption & Infrastructure Spend

AI has moved from pilots to the operating core. In 2024, 78% of organizations reported using AI, up from 55% in 2023; this is a step change that signals enterprise-wide adoption rather than isolated experiments. At the same time, the money is following the momentum: U.S. private AI investment reached $109.1B in 2024, with generative AI attracting $33.9B globally, up 18.7% year over year. Advertising, the business engine for many personalized services, is already predominantly digital.

Keeping AI-driven personalization fresh at this scale is an engineering challenge before it is a modeling exercise. At Meta, Ghosh led the efforts of scaling the ML Platform. Industry ML Models have grown from megabytes size to multi-Gigabytes. The explosion of data and compute hungry models and increase in model sizes were just getting started in 2017. Ghosh’s technical innovations ensured that Facebook’s ML Platform was able to train and host bigger and larger models for inference that were more compute and memory intensive. These critical technical innovations and timely transformation enabled Facebook to scale their AI infrastructure and engage billions of users across multiple apps that fed directly into Facebook’s $100B+ revenue juggernaut. User engagement growth and distribution fuel both advertising and creator monetization which in turn feeds into small and medium business growth and translates into jobs growth for the economy. Meta’s advertising technologies generate $415 billion in economic activity annually for the U.S. and support 3.1 million jobs according to 2022 data.

Ghosh recently spoke at the CED Conference on “Evolving Crafts of Software Engineering with AI Advancements,” sharing this scale-focused approach with engineering leaders. “AI-driven personalization at scale is contrary to perfect predictions: instead, it concerns systems that adapt and stay reliable when demand spikes,” says Ghosh.

Integrity in the Creator Economy

Audiences increasingly discover news and creators inside AI-powered feeds and recommendation systems: 38% of U.S. adults regularly get news on Facebook and 35% on YouTube, whereas 20% do so on Instagram and 20% on TikTok. With reach comes responsibility: in the UK, four in ten adults report encountering misinformation or deepfake content over a four-week period, which is a signal that integrity defenses in AI ranking must evolve as quickly as content formats.

To meet these challenges, Ghosh founded LinkedIn’s Feed Trust team, building it from the ground up and led efforts providing user options to hide posts that users are not interested in, report content that users may find inappropriate or offensive, and launching the platform’s first trust-signal pipelines to make Feed relevant based on these user preferences. These product and technical innovations enabled Trust-aware ML ranking at massive scale. He also authors a career development newsletter that reaches more than 500 subscribers, distilling integrity and experimentation practices for practitioners.

“Trust isn’t built by adding safety after the fact — it’s earned when integrity is woven into the very logic of how a system learns, ranks, and responds.” says Ghosh.

Reliability at Planet Scale

Every millisecond counts when AI inference, features and feedback loops drive billions of feed decisions across ad breaks, launches and viral moments. The business risk is concrete: unplanned downtime costs Global 2000 firms an estimated $400B annually, which is about 9% of profits. When failures do occur, the bill is steep: 54% of organizations report their most significant recent outage cost over $100,000, with roughly one in five above $1M. Regulatory expectations amplify the pressure: under the SEC’s cybersecurity disclosure rules, public companies must file an Item 1.05 Form 8-K within four business days of determining a cyber incident is material.

At LinkedIn, Ghosh led a broad reliability overhaul. He innovated on standardized response frameworks for major incidents across multiple teams which ensured teams could inform and co-ordinate quickly. His work cut incident detection and recovery times, raised service uptime and availability. “Reliability is the quiet hero of great systems — unnoticed when present, unforgettable when absent,” says Ghosh.

The Road Ahead

Industry momentum suggests the stakes will keep rising: global ad revenue is projected at about $1.08T in 2025, even amid forecast downgrades tied to trade uncertainty and de-globalization. As more value shifts into AI-driven experiences, the winners will be those who make personalization, integrity, reliability and compliance work together at the AI infrastructure layer.

Ghosh’s record and contributions map to that playbook and reflect a career spent building systems, growing talent and, better yet, shaping the broader conversation on how AI/Data infrastructure evolves. He is a Scientific Committee member of IEEE ICAH.

“AI infrastructure is the modern railroad — massive upfront spend, but compounding returns for generations.” says Ghosh.