Cimba.ai Inc., a startup making it easier for enterprises to customize language models, launched today with $1.25 million in initial funding.

The pre-seed raise was led by Ripple Ventures with participation from SeaChange and PackVC. The funding round also included contributions from angel investors Chad Sanderson, chief executive of venture-backed data management startup Gable.ai Inc., and Apache Samza creator Chris Riccomini. Samza is a popular open-source tool that developers use to build analytics applications. 

Cimba is led by CEO Subrata Biswas, a former Airbnb Inc. engineering manager who helped build the short-term rental platform’s data management infrastructure. He co-founded the startup with Vishal Das, who was previously an applied scientist at Amazon Web Services Inc. Cimba reportedly has nine employees.

Enterprises often customize the large language models they adopt by training them on internal datasets. A handset maker, for example, might train an LLM on its product documentation to let the model field customer questions about new devices. Depending on factors such as the amount of data involved, customizing an LLM can cost millions of dollars.

Seattle-based Cimba has developed a software platform that promises to ease the process. According to the company, its platform provides the ability to train LLMs on an organization’s internal data using a low-code interface. Cimba says that customized LLMs, or agents, created with its software take only a few days to deploy in production. 

Under the hood, the company is using a collection of off-the-shelf language models to power its platform. Cimba says that the platform leverages both commercial LLMs and models from the open-source ecosystem.

“Most enterprises are scared to adopt AI in their critical business operations as they are required to train or finetune LLMs with their knowledge and data, which can cost millions and be ineffective for their workflow and ROI,” Biswas said. “Creating an easy to use, cost effective platform for businesses to train AI and utilize those in their day-to-day business operation is our mission at Cimba.”

The company allows users to access their customized AI agents through a chatbot interface. According to Cimba, that interface makes it possible to interact with an organization’s internal datasets using natural language commands. A customer service representative, for example, could ask the platform to list unresolved support tickets and organize them by urgency.

Users working with large datasets can optionally have a Cimba-powered agent visualize the results of their queries. A salesperson, for instance, could turn a spreadsheet containing transaction logs into a pie chart that breaks down revenue by product. Moreover, agents created with the platform are capable of performing actions in third-party applications such as data warehouses. 

Cimba’s platform is currently in private beta. The company says that it has several early customers including multiple midsized and large enterprises. 

Image: Unsplash