Streem.ai and Its EIC Accelerator Anomaly Detector: The Evidence

Engineers who test car heaters for thousands of hours collect far more sensor data than they can read. Streem.ai, a Berlin start-up, said its software could read all of it, pick the right algorithm for each data set on its own, and point people to the anomalies that matter. In the March 2020 round of the EIC Accelerator pilot, the European Commission selected the company with a grant-only award, according to the Commission's list of selected companies, which names Streem.ai GmbH in Berlin as the only beneficiary of the project. The EU contributed €1,813,875 to a €2,591,250 project that ran from 1 July 2020 to 30 June 2022, according to CORDIS.

The project ended on schedule, and the company reported the software built. On 19 January 2024 the Charlottenburg district court in Berlin opened insolvency proceedings over the company's assets, which dissolved it under German company law. The proceedings were closed on 20 December 2024 after the final distribution to creditors, and on 24 March 2025 the court registry struck Streem.ai GmbH off the commercial register as a company without assets, according to the register entries reproduced by webvalid. North Data lists the same docket number, 36s IN 6260/23, the December 2024 closure and the start of the deletion procedure in January 2025.

What Is Left of Streem.ai

Nothing operates under the name any more. The company was dissolved when insolvency proceedings opened in January 2024, and since its deletion in March 2025 it no longer exists as a legal entity, according to the register entries on webvalid, which marks it "inaktiv". A list of start-up insolvencies compiled by Arnas Bräutigam on LinkedIn includes Streem.ai GmbH under 23 January 2024, among 33 start-ups it counted for that month. The register entries say the proceedings were lifted "nach Abhalten des Schlusstermins und Vollzug der Schlussverteilung", after the final creditors' meeting and the final distribution, which means there was an estate to distribute; how much creditors recovered is not public.

The website outlived the company. Its homepage, with the EU funding notice and the Berlin address, was still online in a June 2026 copy in the Wayback Machine. Today streem.ai shows a registrar's page saying the domain has "recently registered with namecheap.com". Commercial databases have not caught up: Tracxn still lists the legal entity as active and CB Insights as "Alive".

The Technology: Picking the Algorithm for the Data

Anomaly detection in industry usually works one of two ways. Engineers write rules and thresholds for known faults, or data scientists train supervised models on examples that someone has labelled as normal or faulty. Rules miss faults nobody anticipated, and labels are scarce: most sensor data in manufacturing is unlabelled, as the company's own final report points out.

Streem.ai used unsupervised learning, which looks for data that behaves differently from the rest without needing labelled examples. Its claimed difference was a "benchmarking algorithm" that runs incoming data through several detection methods, "from simple statistical methods to complex neural networks", scores them and picks the best for the data set, according to an October 2022 copy of its website. The CORDIS summary describes the same selection method and promises users "an up to 85% reduction in time spent on data analysis". The pitch was aimed at companies without data scientists: the final report says the tool was "designed specifically for domain experts" and could onboard customers "in less than two weeks". None of these performance figures has been tested independently in any source found for this article; no paper or patent by the company could be found.

The best-documented use was in automotive testing. The company's industry page describes endurance tests of automotive heaters that "run for over 2400 hours and involve 150 different signal inputs", and claims that against the previous rule-based alert system, which analysed 10% of the data, Streem.ai cut daily data inspection time by 85% and found "5 previously unknown anomalies across 9 devices". The heater maker is not named on that page, but the homepage pairs the testimonial above with Webasto's logo, and the CORDIS report names Webasto as an example of the companies the product was for. The CORDIS objective also cites a completed pilot with Roche Diagnostics on analyser machine data, the basis for the project's COVID-19 angle: monitoring diagnostic machines used for virus testing.

What the EIC Project Promised, and What CORDIS Shows

The project's stated aim was "to further develop our product and technology, with a focus on scalability and reliability, and begin early commercialization within the manufacturing sector", according to the periodic report for April 2021 to June 2022, the only report CORDIS publishes. It lists what was built: a dashboard, a microservice backend, self-service data ingestion, a monitoring and logging layer, "7 algorithms" deployed in the machine learning pipeline, the benchmarking algorithm in the backend, and public and private cloud versions. Under the reporting period it lists "Onboarding of early trial customers", without naming any. The only other output on CORDIS is an open research data deliverable, "Provide Pilot Data". No publications or patents are listed.

The commercial promises in the project objective were concrete: 18 new jobs and "cumulative first 5 year profits of €18.84 million with a ROI of 493%". None of that can be found in the public record. The company's website kept showing the logos of Webasto, ZF, Roche, T-Mobile and Daimler under "Trusted by global leaders", but no contract, revenue figure or customer announcement from any of them was found. The 2022 annual accounts were filed and published on 12 August 2024, seven months into the insolvency, according to North Data; their figures are behind a paywall and could not be checked for this article.

1. Scalable product with end-user applicationDeliveredDashboard, microservice backend and self-service ingestion listed in the CORDIS report; self-reported.
2. Benchmarking algorithm in productionDelivered"7 algorithms" deployed and a benchmarking algorithm in the backend, per the same report.
3. Early commercialisation in manufacturingNot achieved"Early trial customers" only; no paying customer or contract announced.
4. 18 new jobsNot achievedCompany insolvent in January 2024 and deleted in March 2025.
5. €18.84M profit in five yearsNot achievedNo revenue figure ever published; insolvency 19 months after the project ended.

The project ran its full 24 months, from 1 July 2020 to 30 June 2022, with no termination recorded on CORDIS, so on the technical side it was delivered on time by the company's own account. The commercial half of the plan was not.

What Happened After the Grant

Outside money beyond the grant is hard to find. CB Insights lists a single incubator or accelerator round with Plug and Play as the investor, and the company's careers page in March 2022 said it was "partnered with Capgemini" and had won "the Plug and Play STARTUP AUTOBAHN Innovation Award", while recruiting a QA working student for "the Streem Engineering team". No priced seed or venture round was found. Tracxn lists the EIC Fund among its investors and two grant rounds of undisclosed size in July 2020 and December 2021. That conflicts with the Commission's list, which records the award as grant only, and with the EIC Fund's official portfolio, whose 16 pages of about 320 invested companies do not include Streem.ai.

The founding date also differs between sources. CB Insights and Tracxn say 2017; the commercial register shows the company entered on 9 April 2018 as Restrunner UG, with Abou George as managing director, and renamed Streem.ai UG in February 2019, when Avraham Elran took over as managing director, according to North Data and webvalid. The team page named Elran as chief executive, George as chief operating officer and Gary Abela, a former City banker and chief commercial officer of the Tech Open Air conference, as chief commercial officer. Elran had earlier founded Plastelina Games, which the CORDIS objective says peaked at 1.5 million unique monthly visitors in 2000; the team page dates that peak to 2001. Apart from project 960719, CORDIS records no other EU project for the company.

Why Europe Should Care

The idea matched a stated EU priority. The Commission's approach to artificial intelligence now includes an Apply AI Strategy, adopted on 8 October 2025, meant to push AI adoption across industrial sectors and above all among small and medium-sized companies. Streem.ai's pitch was exactly that: AI for engineers at manufacturers, such as automotive suppliers, that have the data but not the data scientists. The final CORDIS report framed the market in those terms, citing an estimated EU shortage of 485,000 data workers in 2020.

The case also shows the limits of a grant-only award. The EU paid for most of the development work, the software was built, and then the company needed customers or investors to carry it. The public record shows neither.

The Verdict

Streem.ai's former managers and the EIC have not been asked for comment for this article. It is based entirely on public records, archived company pages and EU project data, linked throughout.