Lamini

Lamini is an enterprise LLM platform for building, tuning, evaluating, and running large language models on proprietary data, with a strong emphasis on secure deployment (cloud VPC to on-premises) and reducing hallucinations via 'memory tuning.' Founded in 2022 by Sharon Zhou and Gregory Diamos and backed by about $25M in funding, it targets engineering teams at data-sensitive enterprises that want to develop LLM capabilities in-house rather than rely solely on external APIs.

Lamini targets enterprises that want to build and operate their own LLMs rather than rely solely on external APIs. The platform provides tooling to fine-tune open and custom models on proprietary data, evaluate them, and serve inference, with an emphasis on running securely in a company's own environment, from cloud VPCs to on-premises and even specific accelerator hardware. This appeals to organizations with strict data-governance or compliance requirements. A notable part of Lamini's pitch is improving factual accuracy on proprietary data. It has promoted 'memory tuning' as a method to reduce hallucinations by tuning models to recall specific facts more reliably, which is important for enterprise use cases like internal knowledge and structured outputs. The platform also supports scaling training and inference across GPUs and nodes. Lamini was founded in 2022 by Sharon Zhou and Gregory Diamos and has raised about $25 million, with a Series A led by Amplify Partners and a notable roster of angel investors and strategic backers including AMD Ventures. Pricing has been published in usage terms (for example, inference and tuning-step rates) but is often sales-led for enterprise engagements, so teams should confirm current specifics directly.