Most AI today is like a black box — you feed it mountains of data and it spits out answers, but nobody can explain why, and if the data is skewed, the answers are too. ALMA built a completely different kind of AI engine based on algebra instead of statistics. Think of it like the difference between guessing the answer from thousands of examples versus actually understanding the rules of the game. The result is AI that can explain its reasoning, respects privacy by learning locally instead of hoovering up all your data, and is far less likely to discriminate against people based on biased training data.
Companies deploying AI face growing regulatory pressure (EU AI Act) to explain automated decisions, prevent discrimination, and protect user privacy — but most current AI is a black box trained on potentially biased data. Businesses in regulated sectors like finance, healthcare, and HR risk fines, lawsuits, and reputational damage when their AI systems make unfair or unexplainable decisions. They need AI that is transparent, fair, and privacy-respecting by design, not as an afterthought.
ALMA built a full software stack for algebraic machine learning: an AML-DL Interpreter, Debugger, and Consistency Checker for developing algebraic models; an AML-IP distributed implementation released as open source; an AML accelerator prototype for high-performance computing; a robotized ironing and garment folding testbed; intelligent tools tested with creative professionals; and open source experimentation tools — 65 deliverables in total.
If you are a bank or insurer dealing with regulatory pressure to explain your AI-driven lending or underwriting decisions — ALMA developed an algebraic machine learning engine that produces explainable, auditable models without depending on massive training datasets. This means your compliance team can actually trace why a loan was approved or denied, reducing legal risk under the EU AI Act.
If you are a garment manufacturer struggling to automate complex handling tasks like fabric folding and ironing — ALMA built and tested a robotized ironing and garment folding testbed that uses algebraic learning instead of traditional AI. The system learns manipulation tasks with less training data and adapts to new fabric types more reliably than statistical approaches.
If you are a design tool company looking to embed AI that creative professionals actually trust and can steer — ALMA developed and tested intelligent tools for supporting creative professionals, built as working prototypes with real designers. These tools let human users interact with and control the AI rather than being handed opaque automated outputs.
ALMA released key components as open source, including the AML-IP implementation and experimentation tools. This means the core technology is freely available for integration. Commercial support or custom development would need to be negotiated directly with the consortium partners.
The project developed an AML accelerator prototype specifically for high-performance processing, and tested a robotized ironing and garment folding testbed at benchmark level. However, full industrial-scale deployment would likely require further engineering beyond the current prototypes.
Core tools including the AML-IP implementation, AML-DL Interpreter, Debugger, and Consistency Checker were released as open source. The project produced 65 deliverables total. Specific licensing terms for commercial use should be discussed with the coordinator, PROYECTOS Y SISTEMAS DE MANTENIMIENTO SL.
Unlike statistical AI (neural networks, deep learning), ALMA's algebraic approach does not fit parameters to data distributions. This means it is far less sensitive to biased or incomplete training data and can integrate hard constraints — like ethical rules or safety requirements — directly into the learning process. The trade-off is that this is a newer technology with a smaller ecosystem.
ALMA was specifically designed around human-centric principles: explainability, bias reduction, privacy-preserving distributed learning, and ethical constraint integration. These align directly with EU AI Act requirements for high-risk AI systems. The project's explainable outputs and auditable reasoning offer a strong compliance foundation.
The project delivered working software including an AML-DL Interpreter (Python/C), an AML-DL Debugger, an AML-DL Consistency Checker, an AML accelerator prototype, the AML-IP distributed implementation, intelligent tools for creative professionals, and open source experimentation tools. All went through multiple release cycles from initial to final versions.
The project ran from September 2020 to February 2025 and produced final versions of all major tools. Open source experimentation tools and documentation are available for immediate evaluation. Based on available project data, a proof-of-concept integration using the open source components could begin relatively quickly, but production deployment timelines depend on the specific use case.
The ALMA consortium of 9 partners across 5 countries (Germany, Spain, Finland, France, Portugal) is research-heavy, with 4 research organizations and 2 universities making up the bulk of the team. The 2 industrial partners (22% industry ratio) and 1 SME — the Spanish coordinator PROYECTOS Y SISTEMAS DE MANTENIMIENTO SL — provide real-world grounding but suggest the project leaned more toward technology development than market preparation. The coordinator being an SME is a positive sign for eventual commercialization, as SMEs tend to be more agile in bringing research outputs to market. For a business considering adoption, the strong research backbone means the science is solid, but you would likely need your own integration effort or a partnership with one of the consortium members to move from prototype to production.
The coordinator is PROYECTOS Y SISTEMAS DE MANTENIMIENTO SL, a Spanish SME. SciTransfer can facilitate an introduction to discuss licensing, integration, or pilot opportunities.
Want to explore how ALMA's explainable AI tools could solve your bias or compliance challenges? SciTransfer can arrange a direct conversation with the development team and help you evaluate fit for your specific use case.
