A data-centric AI company in Mountain View, betting the moat isn't the model - it's the messy pile of documents, tables, and sensor logs you already own.
Here is a fact about enterprise artificial intelligence that everyone in the business knows and almost nobody says out loud: the hard part isn't the model. You can rent a very good model by the token. The hard part is that your company's actual knowledge - the part a competitor can't buy - lives in a 400-page equipment manual, a spreadsheet with merged cells, a decade of maintenance logs, and a scanned invoice that is, technically, a photograph of a number. Corvic AI is a company built entirely around this unglamorous truth.
Founded in 2023 in Mountain View, California, Corvic calls its product the "Intelligence Composition Platform." The pitch, stripped of the capital letters, is that it sits between your raw enterprise data and whatever AI you want to run on top of it, and does the tedious, error-prone work of making that data usable - without a team of engineers hand-building a custom pipeline for every new question you want to ask.
That may sound like a modest thing to build a company on. It is not. The graveyard of enterprise AI projects is mostly full of pilots that died in the data-preparation stage, where the demo worked beautifully on clean sample data and then met the company's real files. Corvic's founders - who came out of Intel, the graph-computing startup Katana Graph, and the distributed-training company Determined AI - had watched this happen enough times to conclude that the problem was worth a company.
Their central bet is philosophical, and it has a name in the industry: they are data-centric rather than model-centric. The model-centric world says intelligence comes from a better model. The data-centric world says most enterprises already have all the raw intelligence they need, badly organized, and the winning move is to fix the data rather than chase the next model. Corvic is wagering that the second story is the more durable business.
What makes this more than a slogan is the specific, slightly nerdy way Corvic handles a document. Most retrieval systems flatten everything into text - a table becomes a wall of words, an image becomes a caption, and a lot of meaning quietly evaporates. Corvic instead treats a PDF as a kind of knowledge graph: titles are titles, paragraphs are paragraphs, images are images, and the relationships between them survive. When the system later goes looking for an answer, it knows that the number in the table belongs to the heading three lines up.
This is the problem in one sentence, and it's worth noting that it's a problem of connection, not intelligence. The reports exist. The databases exist. The sensor readings stream in. What's missing is the connective tissue that lets a machine reason across all of them at once. That tissue is, more or less, the entire product.
Corvic has a habit of trademarking its techniques, which is either good branding or a founder's way of insisting the ideas are real. Either way, three of them do most of the work.
Multimodal embeddings that process data in its native form - tables, relations, text, images, time series - instead of flattening everything into one lossy pile of words.
A framework for producing grounded predictions with traceable origins. In plainer terms: every answer can show its work, which is the difference between "trust me" and "here's why."
Explainable Chain of Adaptive Actions - agent orchestration with a full audit trail, so the AI's reasoning steps are inspectable rather than a black box.
A note on trademarks: the ™ symbols are Corvic's own. Whether the techniques are as novel as the branding suggests is the kind of thing that gets tested in the market, not in a press release.
The abstractions are nice, but the reason a factory or a bank signs a contract is that something specific gets easier. Corvic's public use cases are pleasingly concrete - the sort of jobs that eat a specialist's whole afternoon.
Pipe-and-instrument diagrams are dense, technical, and mostly picture. Corvic pulls structured meaning out of them so an engineer isn't tracing lines by hand.
Life-sciences paperwork is enormous and unforgiving. The platform helps assemble grounded, traceable content from source documents - with the citations intact.
An invoice is often a photo of a number sitting next to other numbers. Corvic turns that pile of images into structured, queryable data.
Field-service technicians ask messy questions of messy manuals. Corvic's retrieval is built to answer them from multimodal source material.
Corvic's founders spent their earlier careers on the deeply technical end of the field - deep learning, graph analytics, distributed machine learning, enterprise infrastructure - at places like Intel, Katana Graph, Determined AI, and Movidius. It shows in the product's fixation on structure and relationships.
Sets the data-centric thesis and does the talking - most of Corvic's public quotes are his.
A distributed-systems and graph-computing background aimed squarely at the retrieval problem.
Owns the product - the trademarked machinery that turns the thesis into something a customer can buy.
Joined in 2026 to run enterprise go-to-market; previously led strategic partnerships at SecurityScorecard and was an operating partner at Intel Capital.
In April 2025 Corvic closed a $12 million seed round co-led by M Ventures - the corporate venture arm of Merck - and Bosch Ventures, with Foothill Ventures, Lam Capital, LDV Partners, K2 Access Fund, and Atlantic Bridge's Brian Long also in. The interesting part isn't the number. It's how Bosch got there.
Bosch didn't invest off a pitch deck. Corvic performed strongly in the Open Bosch GenAI Challenge 2024 - Bosch's own hackathon - and then Bosch Ventures wrote a check. That's roughly the ideal way to raise from a strategic: let the customer watch the technology work on their own problem first.
*Total-raised figure per third-party databases (PitchBook); the disclosed priced round is the $12M seed. Treat the total as approximate.
Merck's corporate venture arm, co-leading the round with a life-sciences lens.
The industrial giant's fund, converted from hackathon judge to investor.
Foothill Ventures, Lam Capital, LDV Partners, K2 Access Fund, and Brian Long (Atlantic Bridge).
Founded in Mountain View by Sabet, Nguyen, and Gill around a data-centric thesis.
Open Bosch GenAI Challenge - a strong showing that would later turn a strategic into an investor.
$12M seed co-led by M Ventures and Bosch Ventures to build out the AI cognitive infrastructure.
Matt Stone named CBO to run enterprise go-to-market.
Corvic V3 goes GA across AWS, Azure, and Google Cloud marketplaces, with new self-serve Individual Plans.
Corvic is competing in the loudest corner of software. Enterprise-AI search and retrieval is crowded with well-funded names - Glean (which raised at a reported multibillion-dollar valuation), Writer, Cohere's North, Vectara, and Dust - plus the cloud platforms' own managed services: AWS Bedrock Knowledge Bases, Azure AI Search, Vertex AI Search. And of course the eternal competitor, the in-house team stitching together LangChain and LlamaIndex themselves.
Corvic's differentiation is less "better search box" and more "better data underneath." The Glean-style pitch is enterprise search across your apps. Corvic's pitch is closer to Palantir's old idea - make the data itself AI-ready and explainable - aimed at industries where a wrong answer is expensive: manufacturing, industrial operations, financial services, life sciences.
The other quietly interesting move is distribution. Landing on all three major cloud marketplaces - AWS, Azure, Google Cloud - and launching self-serve Individual Plans is a bet that the next enterprise buyer might be a single analyst with a credit card, not a procurement committee with a six-month cycle. For an enterprise company, selling to an individual is a small heresy. It may also be where the growth is.
Corvic keeps its product walkthroughs and talks on the web. Start with the source - the company's own site and channels - for the current demo material.
Video availability changes over time; links point to the company's live channels rather than a specific dated clip.
Know a data team drowning in retrieval glue? Send them the story of the company trying to make it go away.
Corvic AI is a Mountain View startup building an Intelligence Composition Platform - the logic layer that turns messy, multi-structured enterprise data (PDFs, tables, sensor logs, images, time series) into reliable, production-ready AI outcomes without bespoke pipeline engineering. Founded in 2023 by engineers from Intel, Katana Graph, and Determined AI, it uses proprietary techniques like Mixture of Spaces embeddings and agentic, auditable orchestration to deliver explainable enterprise intelligence for manufacturing, industrial, financial services, and life sciences customers.
Last updated: