Synthefy, a San Francisco startup building foundation models for structured data, announced on August 18, 2026 it has raised $6.5 million in seed funding led by Wing Venture Capital. The round drew participation from Haystack, Samsung Next, Canonical and Lightscape, along with a cluster of individual backers from OpenAI, Microsoft and Meta.
The funding validates a bet that the next wave of artificial intelligence will extend beyond chatbots and image generators into the numerical tables and time-series data that quietly underpin corporate decision-making. Think pricing algorithms, fraud detection systems, demand forecasting. These are the Excel sheets and database rows that most companies still handle with custom-built models, rebuilt from scratch for each new problem.
Synthefy, founded in 2023, wants to change that. The company trains models on numerical datasets so enterprises can run predictions on new tables without training a fresh model for each use case. Co-founder and CEO Somi Agarwal, who previously worked at Uber's self-driving unit and holds a PhD from the University of Texas at Austin, frames it as infrastructure for problems humans can't solve at scale. She co-founded the company with Sandeep Chinchali, an assistant professor at UT Austin, and Raimi Shah, a principal machine learning manager at Zscaler.
"Each new fraud, pricing or forecasting problem starts again," Agarwal told SiliconANGLE earlier this year. "That work does not compound."
The startup targets retail, financial services, telecommunications, infrastructure, healthcare and defense. According to a press release, the angel investors include Srinivas Narayanan, former CTO of B2B Applications at OpenAI; Aparna Chennapragada, chief product officer of Experiences and Devices at Microsoft; and Manohar Paluri, vice president of AI at Meta. Synthefy declined to disclose its valuation.
In recent months, the company released Nori, an open-source tabular foundation model. Synthefy claims Nori-30M rivals Google's 1.6-billion-parameter TabFM at roughly 2% of the size, and with a "thinking" mode enabled, surpasses it on benchmark tests. The company reported 600,000 downloads and more than 5,000 Python package installs as of its funding announcement, though it hasn't publicly named any paying customers.
The competitive landscape has gotten crowded quickly. Google released TabFM and integrated it into BigQuery. Other players include PriorLabs' TabPFN-3 and SODA-INRIA's TabICL v2. For time-series forecasting specifically, Google's TimesFM and Nixtla's TimeGPT offer adjacent approaches.
Nori's weights and training code are available on Hugging Face and GitHub. Synthefy also offers a hosted API through Baseten's Frontier Gateway and lists an on-premises deployment option on AWS Marketplace. The company said it has integrations with Snowflake, Databricks and Google Cloud, the connective tissue needed to reach data wherever enterprises keep it.
Synthefy will use the capital to expand deployment across production environments and build partnerships with cloud data platforms. Wing's Gaurav Garg wrote in a blog post that the startup already has customer traction across retail, finance, observability and defense, though details remain sparse.
"The rise of foundation models for structured data represents the next major expansion of artificial intelligence," Garg said in the release.
According to its LinkedIn page, Synthefy lists between two and ten employees. Chinchali said in the press release that the vision is "a single model layer for the world's structured data." Whether that ambition can survive contact with enterprise procurement cycles and entrenched vendors remains an open question, but the investor roster suggests some experienced hands believe the opportunity is real.
