Vectara offers an end-to-end RAG-as-a-service platform that embeds generative AI features into applications while asserting it never trains on customer data. The company has developed Mockingbird, a fine-tuned LLM optimized for RAG that Vectara says reduces hallucinations, improves structured outputs, and runs with low latency and cost efficiency. Vectara couples Mockingbird with its open-source Hughes Hallucination Evaluation Model (HHEM) and leaderboard, which thousands reportedly use to mitigate hallucinations. The product positioning targets regulated sectors such as health, legal, finance, and manufacturing where accuracy, security, and explainability are critical. With the new model and tooling, Vectara aims to support integrations with downstream systems and functional calls for autonomous agents. The company plans to ramp go-to-market resources and expand its offering in Australia and EMEA. Vectara was founded by three former Googlers to build a conversational search platform for enterprises that lets developers retrieve and summarize text from large corporate data stores. The company offers an AI-powered, API-based search that can handle queries of arbitrary length, ambiguity and language and returns summaries with citations. Its platform ingests documents in real time and supports client-configurable data retention so organizations can discard originals after indexing to limit residual data. Vectara says it grounds summarization by constraining models to customer-indexed data while training on publicly licensed sources such as Wikipedia. Customers include Conversica, Apex Chat, Metus and Oncotelic Therapeutics, and the company competes with cognitive search products like Amazon Kendra and Microsoft SharePoint Syntex. Vectara has 24 employees and plans to use its recent funding to expand engineering and go-to-market teams with emphasis on European, Middle Eastern and Asia Pacific regions. Vectara offers a neural search-as-a-service that converts queries into vector space and indexes an organization’s data to surface results via vector proximity and cross-attentional ranking. The platform encompasses the full data pipeline, ingesting diverse file types (Word, PDF) and understanding document structure rather than just providing a standalone vector database. Founders Amr Awadallah (CEO), Amin Ahmad (CTO) and Tallat Shafaat (chief architect) are former Google staffers and built the company in stealth under the name ZIR AI. Vectara emphasizes transformer-based models and fine-tuning to capture context and improve query relevance. The company launched its neural search offering with $20 million in seed funding and made the technology available upon emerging from stealth. Vectara plans to expand beyond search over time to add services such as recommendations and tools to surface related topics.

Total raised

$74M

Funding rounds

3

Latest round

Series A

Latest activity

Jul 2024