Samaya AI

Samaya AI builds a financial research platform that lets investment teams compare their own research with company filings, earnings transcripts and licensed financial data. When a company issues new guidance, an analyst can ask what changed since last quarter, open the cited passages to check management’s wording and decide which assumptions in the team’s investment memo need another look. The analyst can also assign an agent to revisit that question each time the company files quarterly results.

Maithra Raghu founded Samaya after more than six years studying deep learning systems at Google Brain. By May 2025, Morgan Stanley was a customer. Its Institutional Securities Group partnered with Samaya to help analysts investigate questions using both the bank’s research library and outside sources.

Samaya trained Samaya Sterling to choose how to investigate a financial question. Quarterly results might call for pulling figures directly from a filing, while a recent market event might warrant a web search first. In the second stage of supervised training, a teacher model tried eight strategies for each expert-written question in that stage. The team selected the highest-scoring answer to each question as a training example. Sterling later learned from graded attempts of its own. It gathers the evidence and passes it to a separate model that writes the report.

Samaya tested Sterling on FrontierFinance, its public benchmark of 220 expert-written investment research questions, which it held out of training. In October 2026, the company reported that its base model, NVIDIA’s Nemotron 3.5 Lightning, met 42.5% of the benchmark’s grading criteria. After training, Sterling met 58.9%.