Harsh Singhal Built the AI Systems That Kept Millions of Users Safe Online. Now He Is Doing the Same for Enterprises
Two environments. Two different scales of complexity. One consistent engineering approach. Harsh Singhal’s professional arc from consumer platform safety to enterprise AI governance covers more distance than a job title change suggests. The social media platforms he worked on served hundreds of millions of users across dozens of languages and cultural contexts, and the enterprise environments he works in now serve organizations where a single data governance failure can trigger regulatory consequences, legal liability, and reputational damage that no algorithm can undo. What connects both worlds is the same core conviction: safety and governance only work when they are built into the system, not bolted onto it afterward.
Singhal currently works as a Software Engineer at Glean, the enterprise AI company, where his focus is on data security, AI governance, and applying machine learning to the challenge of making enterprise AI trustworthy at scale. Before Glean, he spent more than 18 years across LinkedIn, Netflix, Adobe, and Koo, building production AI systems in environments where getting safety wrong had direct and visible consequences for real users.
The Consumer Safety Years
At Koo, the Indian multilingual social platform that reached over 60 million users at its peak, Singhal served as Senior Director and Head of Machine Learning from 2021 to November 2023. The platform’s central engineering challenge was content safety across ten Indian languages simultaneously, in a user environment characterized by code-mixing, transliteration, and script variation that made standard moderation tools functionally inadequate. He scaled the machine learning team from three engineers to twenty, built multilingual content moderation systems using fine-tuned large language models, and led the development of KooBERT, an open-source transformer model optimized for Indian-language content across more than 20 languages. The platform’s content recommendation and safety systems, powered by his team, served millions of users daily, and Koo’s multilingual personalization capabilities drew press attention at the time of launch for their technical sophistication in serving vernacular audiences at scale.
Before joining Glean, Singhal worked across LinkedIn, Netflix, Koo, and Adobe, building AI systems in environments where safety, authenticity, and governance had to work in production. Work spanning bot detection and chat categorization shows that his engagement with platform safety as a technical engineering problem long predates the current wave of industry attention.
At Adobe, prior to Koo, his focus was on trust and safety machine learning in large-scale digital systems where content integrity and authenticity carried direct business and legal implications. Early patents from this period, including US10491697B2 on bot detection and US20120130771A1 on chat categorization, show that his engagement with platform safety as a technical engineering problem began well before it became a mainstream industry concern. “The consumer safety work taught me something that transfers directly to enterprise environments,” Singhal said. “When you are moderating content for tens of millions of users, you learn very quickly that rules without systems are not controls. They are aspirations.”
The Shift to Enterprise
Enterprise environments present a different configuration of the same fundamental challenge. Instead of user-generated posts in multiple languages, the data is internal documents, emails, code repositories, HR records, and financial information moving through AI-powered workflows at speeds and volumes that outpace manual oversight. Instead of community guidelines, the governing frameworks are regulatory requirements, legal obligations, and internal data policies with compliance consequences attached.
Singhal’s work at Glean has contributed to sensitive content detection systems that use enterprise graph signals, document permissions, user activity patterns, and contextual classifiers to identify genuinely sensitive information inside unstructured enterprise data. The approach, publicly described by Glean as combining infotype classifiers with document context and enterprise graph data, achieves accuracy rates above 80 percent on unstructured content, a result that carries real operational weight in a domain where traditional data loss prevention tools have long struggled with false positive rates that made them difficult for security teams to trust and act on consistently.
His recent patent filings extend this work into newer problem areas. US20250371085A1 covers enterprise-aware data security posture management using contextualized access intelligence, while provisional applications address security assurance for AI agents operating in enterprise environments and adaptive enterprise intelligence systems with memory and planning capabilities.
His more recent technical work extends these ideas into a new set of enterprise challenges. It includes enterprise-aware data security posture management using contextualized access intelligence, along with approaches for security assurance for AI agents operating in enterprise environments and adaptive enterprise intelligence systems that incorporate memory and planning capabilities.
“The underlying challenge is the same whether you are moderating a social platform or governing enterprise data,” Singhal said. “You have a policy that says something should or should not happen. You have a system that either enforces that policy reliably or does not. My work has consistently been about building the system that does.”
Why the Continuity Matters
As organizations accelerate AI deployment across functions, the governance gap that Singhal has spent his career addressing is drawing increasing regulatory attention globally, with the EU AI Act, US executive guidance, and equivalent frameworks across Asia-Pacific all pushing organizations toward demonstrable technical controls rather than written commitments. The engineers who spent years building safety and governance systems in demanding consumer environments are arriving at the enterprise frontier with experience that most enterprise-focused engineers are only beginning to acquire. Singhal’s career places him squarely in that group, and the systems he built for 60 million social media users now inform the approach he brings to the governance challenges that enterprises are only beginning to confront at scale.
