Casos de uso

Fábrica IA BSC

Plataformas utilizadas: Deucalion e/ou MareNostrum 5

BlueOasis

Agriculture, Climate & Blue Economy

• Title: Advancing marine monitoring through edge-optimised deep learning: Hydrotwin’s approach to underwater acoustic surveillance
• Short description: blueOASIS is the agile tech partner dedicated to making actionable ocean intelligence more accessible, faster, and more powerful than ever. We are taking deep-tech solutions out of the labs to replace costly, slow and inefficient traditional ocean monitoring.  We’ve built a global network of partners to deploy the integrated technology to finally achieve maritime security and environmental sustainability. We are proving that ocean protection is not just possible, but an inevitable outcome of the new wave of ocean technology.  At the core of everything is Hydrotwin, our AI-powered system that provides real-time underwater intelligence. It delivers acoustic monitoring and metocean data, enabling the creation of real digital twins of the ocean. With its edge processing and AI capabilities, it provides ocean intelligence from anywhere on Earth.
• AI techniques used: deep learning, edge compute, HPC, acoustics signal processing
• Outcomes: -Faster processing pipeline achieved through architecture and infrastructure improvements developed with BSC AI Factory compute support, cutting model iteration time; – Real-time vessel and cetacean detection models trained with the BSC AI Factory runs on edge hardware, removing dependency on costly, delayed manual acoustic analysis. – Scalable groundwork for a global underwater sensor network, supporting environmental monitoring and maritime security use cases across multiple deployment sites
• URL: https://www.blueoasis.tech/

Comudel

Legal & Finance

• Short description: Comudel is a certified accounting firm and an AI-first operating system for small companies. Small businesses lose weeks every year to a manual paper chase: invoices arriving by email, WhatsApp and photo, expenses classified by hand, bank statements reconciled line by line, and accounts closed months after the events they describe. The founder learns how the company is doing far too late to act on it. We built a document pipeline that removes that work. Documents arrive through any channel, are read by OCR and layout models, classified against the Portuguese chart of accounts, matched to bank transactions, and posted as accounting entries automatically. A second layer, a language model adapted to the local tax and labour legislation, answers the questions founders actually ask (what do I owe, what can I deduct, can I afford this hire) and flags the entries a certified accountant should review. Support from the BSC AI Factory gives us the compute and the expertise to train and evaluate these models on documents and legislation, instead of depending on general-purpose models that underperform on local formats and rules.
• AI techniques used: Optical character recognition and document layout analysis (deep learning, computer vision); supervised document and expense classification; named entity recognition and information extraction (NLP); fine-tuning of open-source large language models on tax and labour legislation; retrieval-augmented generation for grounded question answering; machine learning matching for automatic bank reconciliation.
• Outcomes: Automatic processing of accounting documents end to end, from inbound capture to posted entry, cutting the manual handling time per document and removing the monthly document chase for client companies. Accounting closed in real time rather than months in arrears, giving small business owners a current financial position and a cash flow forecast they can act on.
• URL: https://comudel.com/

Forecast Factor

Energy

• Title: ForecastFactor BigData Phase1
• Short description: Companies that buy raw materials (energy, metals, agricultural commodities) make purchasing decisions worth millions of euros every year. Get the timing wrong, and costs spiral. Yet most procurement teams still rely on spreadsheets, gut instinct, or expensive consultants to decide when to buy. Forecast Factor solves this with an AI-powered forecasting platform built specifically for procurement teams. Instead of just showing raw price data like traditional providers, we give a clear, actionable signal: buy now, wait, or hedge. Behind the scenes, our AI engine combines thousands of specialised models (analysing historical prices, market news, and economic indicators) to predict where commodity prices are heading, detect turning points before they happen, and model different market scenarios. The result is a tool that turns complex, unpredictable commodity markets into clear decisions procurement teams can act on with confidence; helping them buy at the right time, avoid overpaying, and manage risk more effectively, without needing a team of analysts or expensive market intelligence subscriptions.
• AI techniques used: Machine learning, deep learning, NLP, ensemble learning, mixture-of-experts architecture, causal inference, time series forecasting.
• Outcomes: Faster model development: scaling from local infrastructure to HPC to train larger ensembles and run deeper backtests, cutting new-commodity onboarding time from months to weeks. Improved forecast accuracy: validated through expanded scenario analysis and multi-year backtesting, strengthening conversion of enterprise pilots into paid contracts. Production-grade scalability: infrastructure and workflows that extend beyond the incubation period, positioning Forecast Factor to scale across additional commodity verticals
• URL: www.forecastfactor.ai

PPA-LeaderTek

OTher

• Title: Scaling Explainable Decision Intelligence
• Short description: Organizations generate massive amounts of unstructured qualitative data such as strategic plans, policy updates, and approval logs that remain entirely siloed. Consequently, critical strategic intent and the underlying rationales behind executive decisions are easily lost in emails and unrecorded threads, creating a severe corporate data paradox. LeaderTek addresses this by transforming an organization’s human resources, strategic plans, and financial goals into a unified system graph where decisions are recorded alongside empirical metrics. The AI-based solution acts as a structured funnel, ingesting messy qualitative data and outputting deterministic, explainable insights. By restricting generative engines strictly to narrative output and keeping all logical reasoning within a transparent, non-black-box architecture, the system guarantees absolute data integrity. This explainable decision intelligence platform allows leaders to act with absolute clarity and complete confidence, providing an auditable institutional memory that is fully compliant with strict regulatory frameworks like the EU AI Act.
• AI techniques used: Explainable AI (XAI), Knowledge Graphs, and Natural Language Processing (NLP). The platform utilizes a deterministic backend to map decision pathways to empirical data rather than relying on black-box assumptions. It builds System Graphs to structure qualitative corporate telemetry, strategic plans, and human resource metrics into a unified, traceable architecture. Natural Language Processing is strictly isolated and used solely for generating narrative explanations of the resulting insights, keeping all logic highly secure and auditable.
• Outcomes: 1. Benchmarking on the Deucalion supercomputer validated high-throughput resilience by successfully processing 921,745 continuous requests over a 40-minute stress-test window with a 0% failure rate and absolute data integrity. 2. Testing on the BSC AI Factory infrastructure proved the platform remains highly responsive under heavy enterprise workloads, sustaining over 200 concurrent users with zero thread timeouts or system slowdowns. 3. The collaborative stress-testing verified the platform’s multi-tenant data barriers and human-in-the-loop design, ensuring full alignment with the transparency mandates of the EU AI Act.
• URL: https://www.ppa-leadertek.com/

Última atualização em agosto 2026