AI LEADERSHIP · GOVERNANCE · TRUST & SAFETY
Leading AI for trust,
safety, and governance.
From consumer data protection at LinkedIn and Netflix to enterprise AI governance at Glean, Harsh Singhal combines hands-on engineering with technical and organizational leadership. At Koo, he led machine learning, personalization, and content moderation for communities across India, Brazil, and other markets.
AI security & governance at Glean
IEEE Senior Member · Forbes Technology Council member
SELECTED LEADERSHIP IMPACT
Developing sensitive-content, access-risk, and AI safety systems that help organizations protect information as they adopt AI.
Enterprise governance ↗Building account-integrity and anti-abuse defenses against account takeover, bots, scraping, and data harvesting.
Consumer trust & integrity ↗Leading personalization and content safety across markets, with Topics discovery that helped users and creators connect in their own languages.
Global platform leadership ↗EXPLORE THE PROFESSIONAL RECORD
01 / PROFESSIONAL MEMBERSHIPS
Elevated to Senior Member of the Institute of Electrical and Electronics Engineers on April 17, 2026.
Professional profileMember since 2026 and contributor, writing about AI security and data protection.
View member profile02 / PUBLISHED PROFILES & MEDIA
Articles about Harsh Singhal and his work, followed by interviews and quoted commentary. Publisher labels are retained; contributed professional writing appears in its own section.
A profile of Harsh Singhal’s approach to multilingual platform safety and responsible AI.
Discusses the connection between consumer content safety and enterprise data governance.
Examines technical approaches to applying governance policy in enterprise AI systems.
Covers Harsh’s work on multilingual moderation, KooBERT, and enterprise sensitive-content detection.
A career profile connecting multilingual machine learning, platform safety, and enterprise AI governance.
A profile covering Harsh’s work on KooBERT and multilingual content moderation at Koo.
Harsh was quoted by Suraksha P on Koo’s use of AI4Bharat’s IndicTrans to support Indian-language content and machine learning. Published in the Bengaluru edition, Techtonic, page 8.
Technology journalist Nick Ellis interviewed Harsh about platform safety, personalization, and extending generative AI features to more languages. Article in Portuguese.
Quoted as Koo’s Head of Machine Learning on the NLP and language models behind multilingual topic discovery.
Coverage of Koo’s Topics launch, with Harsh’s perspective on building NLP capabilities for Indian languages.
Harsh was interviewed by Lisa Weaver-Lambert on Koo’s approach to content relevance and engagement in communities using languages beyond English.
03 / JUDGING & PEER REVIEW
Harsh served as a judge at Rutgers RAISE-26, the Public Informatics Forum on Artificial Intelligence and Society at the Bloustein School. Evaluated student work examining AI’s impact on people and society.
Harsh served as a session chair and peer reviewer for CompSIF 2025 at BMS Institute of Technology & Management, Bengaluru, organized in association with IEEE Bangalore Section. Evaluated research submissions for technical quality, clarity, relevance, and suitability for presentation.
As visiting faculty, Harsh mentored engineering projects, helped define assessment criteria, and judged final project evaluations in AI and generative AI.
04 / GITHUB AI PROJECTS
Harsh develops technical building blocks for AI governance and sensitive-content governance: identifying information at risk, evaluating content against policy, and making model behavior measurable. These are practical concerns for enterprise governance, risk, and compliance (GRC) teams. SPANIEL and safeguard-eval make this work available through his public GitHub repositories.
Explore Harsh Singhal’s GitHub profile ↗
Find the information that needs a decision.
Test what a written rule allows or restricts.
Inspect results, failures, and evidence.
Identify sensitive information before it reaches an AI service.
Harsh Singhal developed SPANIEL’s data preparation, fine-tuning experiments, constrained decoder, and evaluation workflow. The 0.6B-parameter model identifies personally identifiable information (PII) in English text using requested entity types, with local inference and source-preserving annotation.
Explore repository ↗Published model ↗Evaluation & limitations ↗
Turn written content policies into testable model decisions.
Harsh’s OSS-Safeguard Policy Tester lets users select or write a plain-language moderation policy, evaluate an image-generation prompt with gpt-oss-safeguard-20b, and inspect the resulting classification and explanation.
Explore related patents and invention records ↗.
05 / PATENTS & INVENTIONS
Inventions naming Harsh Singhal as a co-inventor. Granted patents and published applications are identified separately.
Co-inventors: Harsh Singhal and Arvind Jain. Filed August 12, 2025; published December 4, 2025. Original assignee: Glean Technologies Inc.
Co-inventor with Heng Wang, Owen S. Vallis, Arun Kejariwal, William Hatzer, and James Koh. Original assignee: Cognant LLC. Priority date: February 15, 2018.
Co-inventor with Pallipuram V. Kannan, Ravi Vijayaraghavan, Rajkumar Dan, and Manish Gupta. Filed June 15, 2011; published May 24, 2012. Original assignee: 24/7 Customer Inc.
Related continuation: US20130211880A1 (application 13/843,226), filed March 15, 2013 and published August 15, 2013. These are related records for the same invention family.
Dates describe historical grants and publications. Follow the source records for subsequent developments.
06 / RESEARCH & PUBLICATIONS
Abhay Bhandarkar, Gaurav Mishra, Khushi Juchani, and Harsh Singhal. Applies topic modeling to LLM conversations to investigate themes and their relationship to user preferences.
arXiv:2510.07557 · DOI: 10.48550/arXiv.2510.07557. Listed as a preprint; no peer-reviewed journal publication is asserted.
Harsh Singhal co-authored this paper with Andrew Rodriguez, W. Art Chaovalitwongse, Zhe Liang, and Hoang Pham.
IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews), volume 40, issue 3, pages 319–329, May 2010.
DOI: 10.1109/TSMCC.2010.2040079
Master of Science in Industrial & Systems Engineering, 2008. Thesis: A New Framework of Optimizing Keyword Weights in Text Categorization and Record Querying.
Rutgers thesis repositoryBachelor of Engineering in Industrial Engineering & Management.
07 / PROFESSIONAL JOURNEY
Across Glean, Netflix, LinkedIn, Adobe, and Koo, Harsh Singhal has applied AI and machine learning to protect users, strengthen product integrity, and govern enterprise risk. His work combines technical depth with responsibility for how systems affect people and businesses.
The journey spans consumer account protection, multilingual personalization and moderation, and enterprise information security. It includes setting technical direction, working across engineering, product, security, legal, and operations, and building the models and systems that put that direction into practice.
SELECTED ROLES & CONTRIBUTIONS
AI security & governance · Software Engineer
Provides technical leadership through hands-on development of AI security and governance systems. His work addresses sensitive-document exposure, enterprise access risks, and the reliability of security decisions as AI reaches more of an organization’s information.
Builds across the full path from models to operational controls: fine-tuning language models for security tasks, improving detection and triage workflows, and evaluating prompt injection, tool safety, and agent behavior. The work helps security teams prioritize real exposure and supports safer enterprise AI adoption.
Sensitive-content engineering ↗Governance patent application ↗Employer profile ↗
Head of Machine Learning & AI
Led machine learning, personalization, and content moderation across a social platform serving users in India, Brazil, and other international markets. His remit connected what people discovered—recommendations, feed ranking, semantic search, and Topics—with the systems protecting them from spam, impersonation, and harmful content.
Led the development of Topics across ten languages used in India, helping people find relevant conversations and creators reach audiences in their own languages. Koo’s August 2022 launch announcement described it as the first social platform to offer Topics across those languages.
Combined product leadership with direct technical work, including collaboration with IIT Madras’s AI4Bharat team on KooBERT and multilingual language-model fine-tuning for toxicity detection. Worked with product, legal, operations, and company leadership to advance platform safety.
Topics launch announcement ↗Brazilian interview on safety ↗International platform context ↗
Data Science & Engineering · Technical lead
Led technical work on account integrity, using machine learning to identify malicious login activity and reduce account takeover. The focus was protecting consumers from unauthorized access to their accounts and the data within them.
Connected risk models with the analytics, data pipelines, deployment tools, and outcome monitoring needed to operate account protection in a global consumer service.
Senior Data Scientist / Data Science
Built machine-learning defenses against automated scraping, bots, payment fraud, and fake job postings. This work protected member information from data harvesting and helped preserve trust in the interactions and opportunities offered on the platform.
Across two tenures, combined hands-on model development with cross-functional work spanning data science, product, legal, and AI. Developed methods to measure anti-abuse systems and understand their effect on the risks facing members’ data.
Senior Engineering Manager
Led ML engineering for fraudulent and non-compliant product usage, linking detection to business decisions. The work included propensity models for conversion to paid usage and graph-based exploration of account-sharing rings and team-license misuse.
Owned delivery across data and model pipelines, fraud-pattern analysis, and campaign targeting—turning technical signals into actions that supported product integrity and revenue protection.
Principal Data Scientist
Built a product-recommendation model that gave sales teams explanations and relevant talking points. Combined account context, sales guidance, and product documentation to turn machine-learning output into decisions people could use.
Head of Data Science & Data Engineering
Established the data science and data engineering organization across the Bay Area and Bengaluru. Architected the analytics and model platform, improved bidding decisions through experimentation, and developed more efficient streaming analytics.
Connected technical architecture with business outcomes: faster access to insights, better-informed advertising decisions, and lower infrastructure costs.
Developed in-app bot detection and models to identify unusual database access, applying machine learning to abuse prevention and operational security.
Turned customer-conversation data into tools for agent development, built the scoring system in Python, and led the analytics team’s adoption of R.
Developed marketing-mix models and an industry-focused R curriculum, combining applied analytics with the work of helping practitioners adopt new technical capabilities.
TECHNICAL DEPTH · ORGANIZATIONAL LEADERSHIP · COMMUNITY
Harsh’s practice extends from writing models and evaluation pipelines to leading multidisciplinary AI work and sharing that experience with others. Public projects such as SPANIEL and safeguard-eval make his governance research inspectable. Academic appointments, student mentorship, DataScience.fm, and practical teaching carry those ideas into the wider practitioner community.
Explore technical work ↗Academic service & mentorship ↗Community & learning resources ↗
08 / ACADEMIC APPOINTMENTS
Alongside industry work, Harsh contributes to AI education through academic appointments, student mentorship, and curriculum advice.
Appointed to the Department of Artificial Intelligence and Machine Learning for a term from November 2, 2025 to November 1, 2026. The role includes capstone guidance and evaluation, curriculum advice, and faculty mentorship.
Nominated by the Governing Body to the Academic Council in the Alumni and Industry Expert category.
Mentored final-year engineering students on AI, generative AI, and large language model projects. Helped define project scope and evaluation criteria and served as a judge for final project evaluations.
09 / APPLIED AI PRODUCTS
Harsh Singhal develops AI applications that help people explore complex information. These projects apply search, language models, and data engineering to chemistry, financial research, and consumer fraud intelligence.
Developed by Harsh Singhal
A molecule and patent search product for chemistry research. Users can search by a molecule’s SMILES representation or drug name to explore similar structures and related patent information.
Harsh’s development work brings together molecular fingerprints, RDKit, database search, and language-model summaries of patents.
Harsh Singhal · Chief Architect
An AI application for exploring public-company earnings transcripts. It combines readable transcripts, highlights, and conversational questions with historical and cross-company research.
As Chief Architect, Harsh connects language-model capabilities to the practical task of understanding company disclosures.
Created by Harsh Singhal
A consumer fraud intelligence product that brings developments across the industry into context. DenseLayer organizes signals by risk category and the roles of stakeholders involved, helping professionals understand where threats, capabilities, and responses fit.
Its industry scanner serves people building risk-reduction and detection systems, with a particular focus on AI risk. The aim is to connect emerging patterns with the decisions and responsibilities of the professionals addressing them.
10 / COMMUNITY & PRACTITIONER EDUCATION
Alongside building AI systems, Harsh shares practical knowledge through a collaborative publication, an online ebook, and video demonstrations. These resources help students and practitioners explore ideas and build working projects.
A place for practitioners to learn—and contribute.
Established by Harsh Singhal, DataScience.fm grew from his personal blog into a publication featuring professionals, researchers, and students. It connects AI and data science concepts with hands-on tutorials, project walkthroughs, and perspectives from its contributors.
Its public archive includes contributor writing from 2021 onward, alongside practical work on SQL, Python, machine learning, and modern AI applications.
Harsh Singhal’s online ebook draws on his experience building side projects from scratch. It combines technical notes, example code, and reflections on turning data into a usable product.
Topics connect the user interface, web services, and data layer with tools such as Python, Docker, language models, and embeddings.
On his YouTube channel, Harsh shares videos about using AI and building with data. The demonstrations complement his written tutorials by showing an application in use.
11 / AUTHORED WRITING
Technical and professional articles authored or co-authored by Harsh Singhal.
Harsh Singhal, Sarika Mohapatra, and Sunil Agrawal describe Glean’s approach to sensitive-content detection in enterprise data. The performance figure is Glean’s reported result for its own systems.
A perspective on making data protection decisions useful for security teams.
Lessons from earlier generations of cybersecurity for securing AI systems.
Explore DataScience.fm and Harsh’s online ebook for more practical learning resources.
12 / INVITED SPEAKING
Listed speaker for Rutgers’ AI Strategy Webinar on how AI enables individuals to build solutions that previously required substantial resources.
Invited by Analytics India Magazine to speak at the Machine Learning Developers Summit in Bengaluru about building data products.
Speaker at the “AIML The Electricity of The Future” seminar, discussing industry applications and learning pathways for engineering students.
CONNECT & CONTINUE EXPLORING
Connect with Harsh about leadership opportunities in AI governance, trust and safety, and risk reduction—or collaboration on research, products, and practitioner education.
LinkedIn ↗GitHub ↗Google Scholar ↗Glean ↗DataScience.fm ↗YouTube ↗