Lamini, a Palo Alto-based startup, has secured $25 million in funding from investors, including renowned Stanford computer science professor Andrew Ng. The company aims to help enterprises deploy generative AI technology effectively. Co-founded by Sharon Zhou and Greg Diamos, Lamini distinguishes itself from other generative AI platforms by focusing on the needs of corporations and delivering high accuracy and scalability.
Many organizations have expressed frustration with the obstacles they face when trying to adopt generative AI across their business functions. A poll conducted by MIT Insights revealed that only 9% of organizations have widely adopted generative AI, despite 75% having experimented with it. The top hurdles include a lack of IT infrastructure and capabilities, poor governance structures, insufficient skills, high implementation costs, and security concerns.
To address these challenges, Lamini claims to have optimized every aspect of its tech stack for enterprise-scale generative AI workloads. This includes hardware, software, and model orchestration. One unique technique Lamini employs is “memory tuning,” which trains a model on proprietary data to ensure high precision and recall of key information.
While the concept of memory tuning is intriguing, it appears to be more of a marketing term than an academic one. There is no substantial research on this technique, making it difficult to validate its effectiveness. Lamini will need to provide evidence of its superiority compared to other hallucination-reducing techniques.
Nevertheless, Lamini has other differentiating factors that set it apart from competitors. The platform can operate in highly secured environments, including air-gapped ones, and offers flexibility in terms of on-premises data centers or public and private clouds. It can also scale workloads elastically, reaching over 1,000 GPUs if necessary.
The co-founders’ strong industry connections have helped Lamini secure investments from prominent figures such as Figma CEO Dylan Field, Dropbox CEO Drew Houston, OpenAI co-founder Andrej Karpathy, and LVMH CEO Bernard Arnault. The company has also received support from AMD Ventures, despite Diamos’ background with Nvidia.
Lamini claims that its model training and running performance is on par with Nvidia GPUs, but this has yet to be independently verified. The company has raised $25 million so far, primarily for expanding its team and compute infrastructure.
While there are enterprise-oriented generative AI vendors in the market, including tech giants like Google, AWS, and Microsoft, Lamini aims to address the specific needs of enterprises seriously. Amplify general partner Mike Dauber believes that Lamini is unique in its dedication to helping enterprises unlock the value of their private data while meeting compliance and security requirements.
Although Lamini has experienced significant growth and has a roster of early customers, including AMD, AngelList, NordicTrack, and undisclosed government agencies, its CEO, Sharon Zhou, remains cautious about revealing specific details. The company is focused on handling the high demand for its services and positioning itself as a reliable and profitable tech company within the hyped AI world.