Mafer AI, a Barcelona-based startup building an AI operating system for R&D teams in formulation industries, has closed a €2 million pre-Seed funding round backed by Kfund, 4Founders Capital, Masia and Lavanda Ventures, the startup investment arm of the Puig family.
It has also secured backing from leading business angels, including Adrián Mato (Andreessen Horowitz scout and GitHub Copilot executive), Manuel Roca (founder of Atrápalo and investor in Wikiloc), Dídac Lee, Fernando Castiñeiras (managing partner at Athos Capital), and Juanjo Mostazo, alongside investors from the cosmetics and fragrance industries and professionals from MIT, Google, Samaipata, Bain, BCG, Shakers and Happy Robot.
Fernando Oliver Jané, CEO and co-founder of Mafer AI, said, “Formulation industries have spent decades accumulating a silent asset, their technical R&D history, without having any way to exploit it at scale. The current generation of AI changes that equation completely: today, using the data of a single customer, we can train proprietary models that radically accelerate the pace at which that company brings formulas to market.
“We are building a new category from Europe, during a window that will remain open for the next three years, and that we believe will not reopen again. With €2 million, an extraordinary team and the support of the Barcelona Supercomputing Centre behind us, we intend to build it ourselves.”
Founded in Barcelona in late 2025 by Fernando Oliver Jané (CEO) and Marc Montalbo Burges (CTO), Mafer AI is the company behind MaferOS, an AI-native operating system for R&D teams across formulation industries, including specialty chemicals, food, beverages, cosmetics, personal care, and fragrances and flavours.
According to the company, it is targeting a silent but structural bottleneck affecting formulation-based industries, which encompasses a broad range of sectors, including flavours and ingredients for food and beverages, fragrances for cosmetics and personal care, as well as pharma and nutraceuticals.
It states that companies operating in these sectors have accumulated decades of technical know-how in their laboratories on how successful products are built, in the form of failed formulas, laboratory analyses, regulatory approvals and technical decisions. Yet this asset remains fragmented across closed laboratory instruments, spreadsheets, regulatory PDFs and, most importantly, the minds of senior experts.
Mafer AI points out that this results in the innovation cycle running between five and ten times slower than the competitive pace of the market, with delayed product launches and shrinking margins. It further notes that when senior experts retire or leave, companies can lose up to two years of institutional knowledge. Meanwhile, technical teams continue to operate on decade-old software infrastructures, while reformulating a product still requires manually recalculating regulatory compliance across dozens of jurisdictions.
MaferOS combines artificial intelligence models with a proprietary architecture adapted to each customer, training proprietary models on each company’s historical data while keeping all information protected and isolated.
The company explained that the platform is structured into specialised modules capable of automating technical decisions that are currently handled manually or through fragmented workflows, from laboratory analysis and data structuring to regulatory compliance and formula recommendation.
“What differentiates Mafer’s approach is not simply applying AI to chemistry, but bringing to the sector the full-stack enterprise software model already reshaping other industries: proprietary models trained per customer, AI agents orchestrated on top of structured data layers, and a team of Forward Deployed Engineers embedded directly within client organisations to bring products into production within weeks,” the Spanish startup mentioned in the press release.
It asserts that it is the same operating model popularised by Palantir and recently validated by companies such as Happy Robot and Wonderful in enterprise environments. Mafer claims to be the first company to bring this approach to specialty chemicals and FMCG.
Its platform enables proprietary models trained on each client’s technical history while protecting proprietary data; specialised AI agents that orchestrate complex technical decisions; automatic structuring of fragmented R&D history, including chromatograms, technical sheets, formulas and regulatory documentation; automated regulatory compliance across dozens of jurisdictions; and production deployment within weeks, all through a long-term partnership model designed to keep clients at the technological frontier.
Just six months after launch, Mafer AI reports that it has already secured close to €300k in committed ARR and is targeting €1 million ARR by the end of 2026. The company currently has three production clients in the fragrance and flavours vertical, where it has structured more
than 50,000 technical documents. It also highlights maintaining an active enterprise pipeline across six industry verticals, with proof-of-concept projects underway in four of them. The startup has additionally won three competitive procurement processes against international consulting firms.
Mafer AI has also been selected by the Barcelona Supercomputing Centre for its AI Factory programme (batches 1 and 2), granting the startup access to more than 50,000 GPU hours on NVIDIA H100 infrastructure and the MareNostrum V supercomputer.
The company plans to use this capital to expand the technical team by hiring additional founding engineers and strengthening both the Forward Deployed Engineering and research functions. The team currently consists of eight people, including the two co-founders, and plans to scale to twelve employees during the second quarter of 2026.
It will also deploy this capital to deepen relationships with existing customers and turn them into flagship public references while expanding new modules on top of the deployed platform, and advance the company’s roadmap toward proprietary foundational models trained at the Barcelona Supercomputing Centre.
