Misam Abbas

Staff AI Engineer, LinkedIn

Misam Abbas is a Staff AI Engineer at LinkedIn currently working on GenAI Safety. With over a decade of experience building large-scale ML systems, he has made significant contributions at Meta and Dropbox. At Meta, he developed user representation models serving billions and created fairness metrics for recommender systems with wide adoption. At Dropbox, he laid the foundations of Dash.ai's AI-powered universal search. A conference speaker and co-author of ranking fairness research, his work eliminates critical bottlenecks to AI adoption while ensuring responsible deployment. He holds an MBA from London Business School and B.Tech in Computer Science from IIT Delhi.

Experience

Staff AI Engineer • July 2024 - Present

As the technical lead for GenAI safety, I ensure the safe deployment of Gen AI across LinkedIn. I lead the architecture and development comprehensive frameworks for real-time LLM moderation and monitoring.

Dropbox

Senior Machine Learning Engineer • May 2023 - July 2024

Team lead for Dash.ai , an AI-powered search and answers system As the ML Tech Lead for Dash.ai, I led the retrieval effort for dash.ai, Dropbox's AI-powered universal search product. I built the search capability from the ground up, architecting a hybrid system combining lexical and vector search. Established the ML flywheel for search, including retrieval, ranking, evaluation, metrics, and experimentations.

Meta (previously Facebook)

Software Engineer(Machine Learning) • October 2018 - September 2022

At Meta, I worked on personalization, and user representation models and fairness in recommender systems , delivering significant, measurable impact on systems affecting billions of users. Led a User Interest Modeling project that enabled a key partner team to achieve a 20% improvement on their primary metric by adapting user representation technology for interest modeling. This helped Facebook show users better interest based recommendations.

Education

Indian Institute of Technology, Delhi

B.Tech Computer Science •

London Business School

MBA, Finance •

Dean's List

Conference Talks & Panels

Panel - Highly regulated environments: optimising generative AI systems in compliance-critical industries

Generative AI Summit • Washington DC • 2025-03-05

Finance and healthcare leaders discussed the technical and regulatory hurdles of launching GenAI while meeting strict data-control and audit standards.

Panel - Explainable AI & today's ethical challenges: balancing innovation, regulation and first-mover advantage

Generative AI Summit • Washington DC • 2025-03-05

Explored transparency, fairness and accountability trade-offs when shipping LLM products under emerging AI-governance regimes.

Ask-Me-Anything with Misam Abbas

AI Beyond The Buzz (AMA) • Virtual (Lu.ma) • 2025-04-03

Wide-ranging Q&A on large-scale ML, LLM safety and career advice for AI practitioners.

Panel - Data-Driven Innovation

SoCal AI Responsibility Summit • UCLA, Los Angeles • 2025-04-19

Student-led discussion on responsible data practices for GenAI.

Panel - Tackling hallucinations, drift & data decay at scale

LLM Ops Summit • Virtual • 2025-05-29

Shared strategies to detect prompt and embedding drift, and to design feedback loops that surface real-world hallucinations before they reach end users.

Talk - LLM System Evaluation the only moat in the GenAI era

Chief AI Officers Summit • New York • 2025-06-05

Presented a framework for objective-driven evaluation pipelines, mixing golden sets and automated scoring to accelerate model iteration.

Talk - Building Trust Through Explainable and Auditable AI

Western New England University • Springfield, MA • 2025-08-28

Explaining Model Decisions both globally and locally using capabilities like SHAP values Trust in Generative AI by focussing on logprobs and model's chain of thought and reasoning traces

Panel Discussion on AI Applications, Compliance, & Ethics

Western New England University • Springfield, MA • 2025-08-28

Participated in a group discussion on AI compliance and ethics and addressed questions on balancing the need for explanations and transparency with the performance of the AI models

Publications

Matched Pair Calibration for Ranking Fairness

arXiv (CoRR) • Research Article • 2023-11-30

Introduces matched-pair calibration, a test that measures subgroup-level exposure bias in score-based ranking systems by comparing outcomes of near-identical item pairs.

A smarter way to evaluate LLM applications

LeadDev • Technical Article • 2025-07-30

A structured evaluation framework for LLM systems that emphasizes defining clear objectives, using curated datasets pre-deployment, and ongoing quality assessment post-deployment via feedback and monitoring.

Computer Method and System for Equity Financing by Retail Investors with Collective Due Diligence Funding

US Patent 7,827,081 • Patent • 2010-11-02

Describes a platform that lets retail investors pool commitments and due diligence costs to acquire private-equity stakes in startups.

Do Computers Understand our Emotions

Blog on Gale, a Cengage Company • • 2018-03-27

An introduction to sentiment analysis that makes the case that computers don't actually understand emotions, but they can make educated guesses about sentiment by analyzing text using computational methods like sentiment analysis. While these tools can process vast amounts of text quickly and identify general positive or negative sentiment, they still fall short of human-level performance and struggle with nuances like sarcasm.