SigOpt builds software that automatically optimizes hyperparameters in machine learning systems, replacing manual tuning by data scientists. Its system observes model performance and suggests new hyperparameters without viewing model inputs or outputs, allowing customers to keep data private. The software can be deployed in private datacenters or private clouds, which the company says differentiates it from competitors such as AWS's SageMaker. SigOpt already counts customers including Hotwire and the Massachusetts Institute of Technology, and CEO/cofounder Scott Clark said the company has made deals with algorithmic trading firms. Financially, SigOpt closed a $6.6 million Series A led by Andreessen Horowitz roughly a year and a half before this investment and has since taken an undisclosed stake investment from In-Q-Tel. The partnership with In-Q-Tel also includes In-Q-Tel selling SigOpt's technology to U.S. intelligence agencies. SigOpt offers a platform designed to continually optimize models and business processes by recommending what to test next rather than just selecting from a few variants. The technology is used to optimize digital assets like ads and websites as well as physical products — one early customer used it to test different chemical combinations for shaving cream. The company targets customers with sophisticated production models or processes that affect business value, including firms in consumer packaged goods and financial services. SigOpt's product does not require a team of expert data scientists to tune models, according to CEO and co-founder Scott Clark. Over time the company plans to offer more of a self-serve product. The business recently completed a Series A financing, signaling early investor confidence. SigOpt offers a platform to "optimize everything," aiming to help customers improve outcomes across contexts such as ad campaigns, machine‑learning workloads, and physical experiments. The product can ingest experiment inputs and results (manually for physical tests) and recommend the next variations to test. Its core technology is the Metric Optimization Engine (MOE), which was developed by co‑founder Scott Clark and is available as open source; SigOpt is building premium products and services around the MOE. The system is designed to identify "points of highest improvement" while balancing exploration and exploitation. Customers cited in the article include Pembient, which is using SigOpt to develop a synthetic rhino horn. The company recently raised seed funding to hire its first engineers and move into its first San Francisco office.

Total raised

$9M

Funding rounds

3

Latest round

Series A

Latest activity

Aug 2016