Predibase
Predibase built a developer platform to make state-of-the-art machine learning and generative AI easy to develop, fine-tune, deploy, and operationalize. Its focus was helping organizations move from AI prototypes to production with open-source models, declarative workflows, and cost-efficient inference.
At a Glance
- Enterprise AI and data-science teams
- Software developers and ML engineers
- Fortune 500 organizations
- Startups building specialized AI products
- +1 more
AI Tools by Predibase
(1)
LoRAX
LoRA Adapter Inference Server
Discussions
No discussions yet
Be the first to start a discussion about Predibase
Latest News
Rubrik announced an agreement to acquire Predibase to accelerate enterprise and agentic AI adoption; terms were undisclosed.
Predibase launched an end-to-end reinforcement fine-tuning platform for LLMs.
Predibase launched its next-generation Inference Engine for faster, cost-effective small-language-model serving.
Predibase launched the Fine-Tuning Index ranking open-source LLMs across common production task types.
Products & Services
Low-code/declarative end-to-end ML platform for connecting data, training models, and deploying them through REST endpoints, a Python SDK, or PQL, while retaining fine-grained controls.
Developer platform for fine-tuning and serving open-source LLMs, including parameter-efficient LoRA workflows, model deployment, LudwigGPT, and serverless endpoints that scale with usage.
Apache 2.0 open-source multi-LoRA inference server that dynamically loads and merges adapters, batches requests for heterogeneous adapters, and serves hundreds or thousands of fine-tuned models on shared GPU infrastructure.
Collection and web application demonstrating specialized fine-tuned LLMs for tasks such as sentiment analysis, summarization, translation, intent classification, and generation.
Market Position
Predibase positioned itself between heavyweight cloud ML platforms from AWS, Google, and Microsoft and simpler but more constrained low-code/no-code tools. It differentiated through open-source model support, declarative workflows, developer escape hatches, parameter-efficient fine-tuning, and serving many specialized adapters on shared GPUs; adjacent competitors included managed model platforms and inference providers such as Hugging Face, Replicate, Modal, Anyscale, and Baseten.
Leadership
Founders
Devvret Rishi
Former Google product leader and the first Google AI product manager for Kaggle; he also worked on large-scale ML infrastructure and studied computer science and AI at Harvard. He was Predibase co-founder and CEO.
Piero Molino
Former Uber AI research scientist and creator of Ludwig, an open-source declarative deep-learning framework; previously worked at IBM Watson and founded QuestionCube. He was Predibase co-founder and later chief science officer.
Travis Addair
Former Uber AI/software engineer who led the team behind Horovod, the distributed deep-learning framework. He is also a lead maintainer of LoRAX and Horovod and a core Ludwig contributor; he was Predibase co-founder and CTO.
Chris Ré
Stanford professor and creator of Overton, a proprietary ML system developed at Apple; he was part of the founding team and an early technical architect/advisor.
Executive Team
Devvret Rishi
Co-founder and CEO (later GM of AI at Rubrik)
Former Google AI product leader and first Google AI PM for Kaggle; background in open-source ML infrastructure and Harvard computer science/AI.
Travis Addair
Co-founder and CTO
Former Uber AI engineer and leader of the Horovod team; distributed ML, scalable training, and open-source AI systems specialist.
Founding Story
The founders experienced first-hand at Uber, Google, and Apple that bringing ML models into production could take months or a year. They created Predibase to productize the declarative approach behind Ludwig and related internal platforms, letting users specify what they wanted their models to do while the system handled much of the implementation and infrastructure.
Business Model
Revenue Model
Cloud and enterprise software services for fine-tuning and serving models, including usage-based/serverless inference, managed training, and private-cloud or VPC deployments. The platform emphasized paying for actual endpoint usage rather than dedicated GPUs; enterprise terms were sales-led.
Pricing Tiers
A free two-week trial was announced with general availability in 2023.
A free developer tier was offered for experimenting with fine-tuning and serving; paid enterprise pricing was not publicly listed.
Custom commercial deployment for Fortune 500 and other organizations, including private-cloud or VPC operation.
Target Markets
- Enterprise AI and data-science teams
- Software developers and ML engineers
- Fortune 500 organizations
- Startups building specialized AI products
- Organizations in industries requiring private-cloud or controlled data/model deployment
- Enterprise predictive ML and classification
- Fine-tuning specialized LLMs for domain-specific business applications
- Sentiment analysis, summarization, translation, intent classification, and text generation
- Customer-facing and internal AI assistants and chatbots
- Serving many personalized models for different users, departments, or use cases
- Cost-efficient production inference and private-cloud AI deployments
- Paradigm
- Koble.ai
- Enric.ai
- Sekure Payments