The AI data engineering platform that connects to your whole stack, fixes broken pipelines automatically, and tries to make data infrastructure invisible.
Here is a thing that is true about enterprise artificial intelligence, and that almost nobody wants to say at the demo: the model is usually fine. The problem is the data underneath it. You can buy the best model in the world and point it at a warehouse where a column quietly changed meaning three weeks ago, a nightly job failed silently, and the one engineer who understood the join left for a competitor - and the model will confidently produce garbage. This is not a modeling problem. It is a plumbing problem. And plumbing, it turns out, is expensive.
Upriver is a company built on the premise that this plumbing should mostly run itself. Founded in 2024 in Tel Aviv by Ido Bronstein and Omri Lifshitz, it sells what it calls an AI data engineering platform: software that connects to a company's full data environment - Snowflake, Databricks, BigQuery, Airflow, dbt, the whole modern stack - and then does the unglamorous work. It explores the environment. It builds and validates pipelines. It notices when something breaks and traces the break across the six disconnected tools where the evidence is scattered. And it tries to write down the tribal knowledge that normally lives, undocumented and un-backed-up, in the head of one senior engineer.
The pitch, in Bronstein's words, is almost aggressively modest. Not "we will make your data brilliant." Just: we will make it invisible - the way electricity and running water are invisible, which is to say you only think about them when they fail.
"We built Upriver to take that burden off data teams entirely. Our goal is to make data infrastructure invisible."
If you have spent any time near a data team, you know why this is a good sales line. The most expensive hours in enterprise AI are not spent on GPUs. They are spent by a tired engineer at 2 a.m. answering the question "why does this dashboard say we lost money in a country we don't operate in," working backward through a pipeline nobody fully remembers. Upriver's bet is that an agent can do most of that backward-tracing, and that the human, freed from it, gets promoted - from the person who holds the context to what the company calls the knowledge engineer: someone who decides what the data means, rather than someone who spends their day proving it isn't broken.
The two founders are graduates of Talpiot, Israel's most selective military-technology program, and recipients of the Israel Defense Prize. They spent roughly a decade building large-scale intelligence systems - the kind where a data error is not a wrong dashboard but a wrong decision - before deciding the tooling they had improvised internally deserved to be a company.
Holds a master's in computer science from Tel Aviv University (2018-2020) and served roughly six years in Unit 8200 in cyber-operations and data-platform roles. He is the company's public voice on "data observability at the source."
Fellow Talpiot graduate who spent a decade building intelligence systems at scale. He leads the engineering behind Upriver's context engine and coordinated agent system.
The crowded category next door to Upriver is data observability - a decade of tools that are very good at telling you your data is broken, usually with a red number and an alert. Upriver's argument is that watching is not the same as fixing, and that the interesting work is upstream, at the source, before a bad record has a chance to propagate into forty downstream tables. Hence the name.
Connects to the full environment and builds a live map of how data is structured and how it flows - the model of your warehouse that normally only exists in one engineer's memory.
A coordinated system of agents builds and validates pipelines across fragmented environments, catches issues, and maintains pipeline health end to end.
Records the undocumented context - what a field means, why a rule exists - so it stops walking out the door when someone leaves.
Built for Snowflake, Databricks, BigQuery, Airflow and dbt - the stack most enterprise data teams already run on.
Integrates with AI development tools so coding agents can reach into live production data safely instead of guessing.
The end state Upriver is selling: pipelines you never have to think about, and a reliable data foundation AI can run on without constant upkeep.
Product capabilities as described by the company and press coverage; specifics may evolve.
In June 2026 Upriver announced roughly $14M in funding - an initial round of about $4M followed by a $10M round - led by Valley Capital Partners and Hetz Ventures. What is telling is the angel list: Lew Cirne, who founded New Relic; Yotam Segev and Tamar Bar-Ilan of Cyera; and Abe Gong, who built the widely used open-source data-quality tool Great Expectations. When the people who defined adjacent categories write checks, it is a signal about where the category is heading.
| Round | Amount | Date | Lead Investors |
|---|---|---|---|
| Early rounds (reported as Series A in some sources) | ~$14M total | Announced Jun 2026 | Valley Capital Partners, Hetz Ventures |
Sources describe the raise variously as seed and Series A; roughly $4M was raised earlier, followed by a ~$10M round. Angels: Lew Cirne (New Relic), Cyera founders, Great Expectations founder.
For a company barely two years old, the logo wall is notable. Named customers include the game-engine giant Unity, the media group Daily Mail and General Trust (DMGT), and data companies Nimble, Bright Data and Bigabid. Nimble reported a 60% productivity increase after adopting the platform - the kind of number that, if it holds across customers, explains the investor enthusiasm.
"We want to empower data engineers to shift from being the person who holds all the context to becoming the knowledge engineer."
Bronstein and Lifshitz leave a decade in military intelligence to build an AI-native data engineering platform, drawing on the internal tooling they'd improvised.
The context engine and agent system take shape; early enterprise users including Unity and DMGT sign on, and partnerships with Databricks and Snowflake form.
Funding led by Valley Capital Partners and Hetz Ventures - with angels from New Relic, Cyera and Great Expectations - to scale engineering and go-to-market.
Upriver is arriving into a busy 2026. Data observability alone has more than a dozen credible players - Monte Carlo, Metaplane, Anomalo, Bigeye, Soda, Sifflet, Elementary - plus foundational tooling like dbt and Great Expectations. The bet that separates Upriver is one of posture: most of the field observes and alerts; Upriver wants to act, fixing data at the source and closing the loop without a human in the middle. Whether "autonomous data engineering" is a durable category or a feature the incumbents absorb is the open question - and it's the one the $14M is meant to answer.
Compiled from public sources, July 2026. Figures and details are approximate where sources vary.