Predibase

Predibase is a company developing a low-code AI platform for developers. The Predibase platform allows users to train, fine-tune, and deploy any machine learning model. Using a low-code declarative approach to machine learning, Predibase aims to make it easier for engineers, developers, and data scientists to build and optimize AI models while only requiring a few lines of code. Predibase is built on the cloud using open-source technology, such as Ludwig and Horovod. The platform can be used to develop a range of use cases, including unstructured data analytics, recommendation systems, customer service automation, churn prediction, predictive lead scoring, anomaly and fraud detection, and demand forecasting.

Based in San Francisco, Predibase was founded in 2021 by Piero Molino (CEO), Travis Addair (CTO), and Devvret Rishi (chief product officer). Molino and Addair previously worked together at Uber, where they saw the time it took to get machine learning models in production. Molino developed Ludwig, a popular open-source framework for creating deep learning models. Addair led the team behind Horovod, an open-source framework for efficiently scaling and distributing deep learning model training to massive amounts of data.

Predibase emerged from stealth in May 2022 with its commercial platform, which had been in beta with Fortune 500 enterprises for nine months. On the same day, the company announced $16.25 Million in Seed and Series A Funding led by Greylock with participation from angel advisors such as Zoubin Ghahramani (Professor of Information Engineering at Cambridge and Sr Director of Research at Google), Anthony Goldbloom (CEO of Kaggle), Ben Hamner (CTO of Kaggle), Remi El-Ouazzane (former COO of Intel AI), Varun Badhwar (former SVP of cloud security at Palo Alto Networks) and Yi Wang (principal engineer at Meta).

On May 31, 2023, Predibase announced a $12.2 million expansion (led by Felicis) to its $16.25 million Series A funding round. The company also stated that its platform was available to all developers. During the beta period between May 2022 and May 2023, the platform was used to train over 250 models.

Predibase offers a range of features to help build, customize, and deploy machine learning pipelines.

• Data connectors—ingest structured or unstructured data or connect to popular data sources in the cloud, such as Snowflake, Databricks, AWS, Azure, and Google Cloud.
• Declarative model building—create models by selecting features and the wanted target to predict. Users can choose from a custom model architecture or let Predibase suggest a series of experiments depending on the task. Users can also leverage popular pre-trained models through integrations with Hugging Face.
• Composable model architecture—evidence-centered design (ECD) breaks down machine learning pipelines into composable parts, such that users can leverage different techniques for each input feature and output type.

• Collaborative model repository—unified view of all model versions associated with a project, track model changes via an intuitive lineage graph, and compare model performance with analytics and visualizations.

• Intelligent recommendations—Predibase's data science copilot offers model tuning suggestions and recommended configurations to improve model performance.

• Predictive Query Editor—PQL editor enables teams to explore models with similar SQL-like commands. Run different scenarios to determine how parameters or features will impact model performance.

• Serverless Managed Compute—distribute model training using scalable cloud infrastructure.
• Model Deployment Options—schedule batch jobs or run low latency, real-time inference with deployments to REST APIs.