Startup Profile: Voyage AI

Voyage AI - Supercharging Search and Retrieval for Unstructured Data

Voyage AI - Supercharging Search and Retrieval for Unstructured Data 

Voyage AI at a Glance

Team: Founded by Stanford computer science professor Tengyu Ma, Voyage AI’s team comprises leading AI researchers and engineers from Stanford, MIT, Berkeley, Princeton, and CMU, who have conducted over five years of cutting-edge research on training embedding models.

Founded: September 2023

Location: Palo Alto, California

Funding: $30M in total in Oct 2024, led by CRV, Wing VC, and Conviction

TLDR: Voyage AI builds best-in-class embedding models and rerankers that boost the quality and efficiency of unstructured data search and retrieval in retrieval augmented generation (RAG). Built by world-class researchers, Voyage AI outperforms in all dimensions — accuracy, latency, and costs — while providing flexible deployment options and licensing. Voyage AI offers general-purpose embedding models, multimodal-embedding models, domain-specific models tailored to fields such as FinanceCode, Legal, and Multilingual and rerankers.

Company Mission: Recognizing the widespread need for proprietary information ingestion in large language models with the retrieval-augmented generation (RAG), Voyage AI focuses on developing best-in-class embedding models and rerankers—critical and innovative AI components that drive retrieval accuracy. Our mission is to elevate retrieval accuracy to a near-perfect level for data spanning diverse modalities and formats.

Founder: Tengyu Ma is an assistant professor of computer science at Stanford. His research interests broadly include topics in machine learning, algorithms, and their theory, such as deep learning, (deep) reinforcement learning, pre-training / foundation models, robustness, non-convex optimization, distributed optimization, and high-dimensional statistics.

Business Model: We offer public APIs with usage-based pricing and provide flexible deployment options. Our solutions are available on all major clouds and data platforms, supporting SaaS, customer-tenant (in-VPC), and on-premise deployments. Additionally, we offer model licensing and fine-tuning to help customers optimize performance for their specific use cases.

Quick Stats: Our recently launched voyage-multimodal-3 improves retrieval accuracy by 19.63% over the next best-performing model. Please find more details about the evaluation results.

Our latest generation of voyage 3 series models outperforms OpenAI-v3-large across all domains, offering 3-6x smaller embedding dimensions and 4-12x smaller model sizes — leading to significant cost savings. Resulting in an 83.5% cost reduction while having a 7.55% increase in quality. Find more details about the evaluation results.

Our latest rerankers, rerank-2 and rerank-2-lite, improve the accuracy by an average of 13.89% and 11.86%, 2.3x and 1.7x the improvement attained by the latest Cohere reranker (English v3), respectively, when applied on top of OpenAI’s v3 large model.

We’re growing fast and looking for passionate individuals to join us! Whether you’re in backend engineering, sales, marketing, or another field, we’d love to hear from you. If you think you’re a great fit, reach out to Voyage today!

If you’re a startup interested in being profiled, please email us at [email protected]