Captain is a managed retrieval-augmented generation (RAG) platform out of Y Combinator's Winter 2026 batch. It lets developers connect data sources - S3, SharePoint, Google Drive, Slack, Notion and more - with a couple of API calls, then handles the full pipeline behind the scenes: OCR, chunking, embedding, vector storage, hybrid search, and re-ranking. The company claims roughly 95% retrieval accuracy on a published benchmark versus about 78% for hand-built pipelines, and positions itself as 'the Snowflake for unstructured data.' In March 2026 it added Odyssey, a proprietary private-market intelligence dataset queryable through the same API.
Contextual AI is a Mountain View enterprise AI company that builds a context engineering platform for production-grade, retrieval-augmented generation (RAG) agents. Founded in 2023 by RAG co-inventor Douwe Kiela and Amanpreet Singh, it packages retrieval, reranking and grounded language models into an end-to-end system that lets regulated enterprises deploy accurate, source-grounded AI on their own documents. The company has raised $100M from backers including Greycroft, Nvidia, Snowflake, HSBC and Bezos Expeditions, and counts Qualcomm and HSBC among its users.
Vectara is a Palo Alto-based enterprise AI company building a trusted, end-to-end platform for retrieval-augmented generation (RAG) and AI agents. Founded in 2022 by former Google AI researchers, it lets companies embed grounded, citation-backed GenAI - conversational assistants, semantic search, and agents - into their products while actively detecting and correcting hallucinations at runtime. Its stack includes the Boomerang embedding model, the RAG-tuned Mockingbird LLM, the open-source Hughes Hallucination Evaluation Model (HHEM), and a Guardian Agent layer, positioning Vectara as a leader in factual, secure, compliance-ready enterprise AI.
LlamaIndex is a San Francisco company building the data framework and cloud platform that lets enterprises turn messy unstructured documents into knowledge agents powered by large language models. Its open-source library is one of the most-used scaffolds for retrieval-augmented generation, and its hosted product, LlamaCloud, packages parsing, extraction, and indexing for production teams.
Harrison Chase is the co-founder and CEO of LangChain, the open-source framework that became the default scaffolding for developers building LLM-powered applications. Starting as an 800-line Python package he wrote as a side project in October 2022, LangChain has grown to serve 1 million+ developers, 80 million monthly downloads, and blue-chip enterprise customers from Uber to JPMorgan - reaching a $1.25 billion valuation after a $125 million Series B in October 2025. A Harvard-trained statistician who got into machine learning through sports analytics, Chase built LangChain to solve the problem he kept running into at work: there was no good way to chain LLM calls together into reliable, production-ready applications.
Jerry Liu is the co-founder and CEO of LlamaIndex, the open-source data framework that became essential infrastructure for connecting large language models to enterprise data. What started as a weekend side project in November 2022 - an indexing tool he built to feed his own data into GPT-3 - grew into a company with $46.5M in total funding, 600,000+ monthly downloads, and clients including Salesforce, KPMG, and Carlyle. A Princeton computer science graduate who published GAN research as an undergrad, Liu moved from Quora's feed ranking team to Uber's autonomous vehicle research labs before co-founding LlamaIndex with former Uber colleague Simon Suo in early 2023.

ZeroEntropy is the AI infrastructure company fixing the broken retrieval layer of modern AI applications. Founded in 2024 by Ghita Houir Alami (CEO) and Nicholas Pipitone (CTO), the San Francisco–based startup builds rerankers, embedding models, and end-to-end search infrastructure that outperforms Google, OpenAI, Cohere, and Voyage on public benchmarks. Backed by Y Combinator (W25) and a $4.2M seed round led by Initialized Capital, ZeroEntropy's products — zerank-2, zembed-1, zsearch, and ze-onprem — are used by enterprises including Assembled (serving Stripe, Canva, Robinhood, and Notion). The company's proprietary zELO training methodology, derived from chess Elo ratings and the Thurstone statistical model, produces models with calibrated relevance judgments that binary labels cannot replicate.