Regular updates on the latest VC-backed AI startups. Follow along to stay informed!
Vectara, a generative AI search platform, raised $28M in Seed funding. Race Capital led the round.
Problem to be Solved
Semantic search (made possible by embedding models) is game-changing for users but difficult to manage for developers. Vectara aims to make semantic search faster, more reliable, and easy to use by abstracting away the details of implementing semantic search from scratch. It is as simple as importing a library and making an API call.
How They Use AI
Vectara is an abstraction of one of LangChain’s most popular use cases: searching and summarizing answers from an internal knowledge base. Although LangChain enables this use case, Vectara forgoes the versatility of LangChain and instead provides an extremely simple API for the common use case of semantic search. As far as we can tell, Vectara’s edge is its middleware (chaining models and processes together) and does not appear to be training any in-house models, although that is likely on its roadmap.
Business Model
Freemium. Developers can use Vectara’s LLM-powered search as a service for free for up to 15,000 queries per month. For applications running at scale, Vectara charges a usage-based subscription fee.
UpCodes, a construction codes platform and database, raised $3.5M in Series A funding. The deal was led by Building Ventures, with participation from CapitalX and Bragiel Bros.
Problem to be Solved
Many of the over five million sections of building code and constantly changing building code amendments in the U.S. are only available in physical reference books. Even when digitized, they are hard to search, making it difficult for home builders and homeowners to stay compliant.
How They Use AI
UpCodes uses a large language model to generate responses to users’ natural language queries based on relevant “building code” sections obtained from a vector database. Called Copilot and built on the foundation model GPT-4, the tool serves as a research assistant, answering complicated code questions and annotating responses with links to relevant sections of code.
Business Model
SaaS. The base version of the product starts at $39 per month when billed annually. UpCodes now has more than 650,000 monthly active users (most came before the AI products as the business was founded in 2016).
Poolside, a developer of AI foundation models pursuing AGI, raised $26M of Seed funding. Redpoint Ventures led the round.
Problem to be Solved
TBD. Founder Jason Warner, an investor at Redpoint and the former CTO of GitHub, says “We are a frontier thing. We are going to be exploring uncharted territory,” and “The only competitor we think about is OpenAI.”
How They Use AI
Poolside will develop proprietary AI models specifically designed for coding and software development. As the CTO of GitHub, Warner oversaw the creation of GitHub Copilot, one of the most revolutionary software development tools of all time (built using OpenAI’s models in partnership with GitHub’s owner, Microsoft). Warner says Poolside’s first product will be a large language model that is targeting a trillion parameters (right now the biggest models are in the hundreds of billions of parameters), and that he hopes to build a “subsystem” that could potentially have GitHub Copilot or Replit’s Ghostwriter as customers.
Business Model
Poolside will presumably combine cutting-edge AI research with commercial applications of AI. Similarly to OpenAI, the company could charge businesses and developers for API access to generative models or make money by selling subscriptions to consumer applications.