Marah Abdin

Also known as Marah I Abdin

Marah Abdin is an AI researcher and team lead whose work connects Microsoft’s Phi language models with Poolside’s Laguna models. She worked at Microsoft Research from 2019 to 2023 on natural language processing, human-centered and applied AI, urban environmental sensing, and statistical and network-science research, then worked across Microsoft Research and Microsoft AI on Phi from 2023 to 2025. Between 2024 and 2025, she coauthored technical reports for Phi-3, Phi-4, and Phi-4-reasoning. Since 2025, she leads Poolside’s synthetic-data team and builds the Laguna language models; at AI Engineer World’s Fair in 2026, she reports that synthetic data accounts for 13% of Laguna XS.2’s pre-training mix.

Now at: Poolside
Links: Website, Google Scholar, X, LinkedIn and GitHub

Timeline

2019–2023

Research at Microsoft Research

From 2019 to 2023, she works in Microsoft Research labs in Redmond and New York on natural language processing, human-centered and applied AI, urban environmental sensing, and statistical and network-science research.
2023–2025

Phi work at Microsoft

From 2023 to 2025, she works as an MTS and researcher across Microsoft Research’s Physics of AGI and AI Frontiers groups and Microsoft AI. She contributes to Phi-1.5 through Phi-4-reasoning, with synthetic-data work focused on scale, quality, diversity, and high-reasoning capabilities in STEM and code.
April 2024

Phi-3 for phones

In April 2024, she coauthors the Phi-3 technical report, which introduces phi-3-mini, a 3.8-billion-parameter language model small enough to be deployed on a phone.
December 2024

The Phi-4 report

In December 2024, she coauthors the Phi-4 technical report, which presents a 14-billion-parameter language model and incorporates synthetic data throughout its training process.
April 2025

Phi-4 reasoning

In April 2025, she coauthors Phi-4-reasoning, a 14-billion-parameter reasoning model that generates detailed reasoning chains.
May 2026

The Laguna M.1/XS.2 report

In May 2026, she coauthors the Laguna M.1/XS.2 technical report, which presents Mixture-of-Experts foundation models built for long-horizon, agentic coding.
2026

Synthetic data at 13%

At AI Engineer World’s Fair in 2026, she discusses Poolside’s synthetic-data pipeline and reports that it accounts for 13% of Laguna XS.2’s pre-training mix.
Since 2025

Synthetic-data lead at Poolside

Since 2025, she leads Poolside’s synthetic-data team across pre-training, post-training, reinforcement learning, and agent-based generation while building the Laguna large language models.