Deutsche Telekom powers Open Models AI Codefest across EMEA with NVIDIA

Over five weeks, September 9 to October 7, 2026, fourteen teams from across EMEA tackled pressing public-sector challenges: from digitizing handwritten Polish hospital records to detecting dark vessels in European waters to building AI tutors for children in Central Asia. Together, they delivered a full-stack, all-European AI story: infrastructure and compute, open models and curated data, a developer ecosystem, and real applications spanning healthcare, public safety, finance, disaster resilience, and government services.

One Codefest. Fourteen teams. Thirteen countries. 

Fourteen teams from across EMEA took part: zehnlab.ai (UAE), Galene.AI (Italy), ABA Technology (Morocco), Disseqt AI (Ireland), Wind (Egypt), Cognideck (United Kingdom), Algoritmica GmbH (Germany), Aivisory (Germany), Saventic Health (Poland), Pleias (France), Bineric AI (Norway), Supertext (Switzerland), Universidad Carlos III de Madrid (Spain), and UB Robotics (Portugal) developed applications and solutions on NVIDIA Nemotron open models and libraries.

Following Demo Day, the jury selected five winning teams: Pleias, zehnlab.ai, ABA Technology, Disseqt AI, and Algoritmica GmbH, who will present at the NVIDIA GTC Berlin Public Sector Forum on October 20.

• Pleias (France): Built SYNTH, a data pipeline that trains small, open medical reasoning models from raw documents using NVIDIA NeMo framework. The first European Engram-enabled model, highly specialized in complex medical knowledge and fluent in the 3 languages spoken in Belgium and English. The model outperforms larger models on multilingual medical benchmarks.
• zehnlab.ai (UAE): The team extends Soro, Tajikistan’s national LLM, into an AI agent that makes correct decisions, calls tools, and searches the web in Tajik using the NVIDIA Nemotron open datasets. The Agentic Soro is the next release for 4,000+ schools and 70+ state bodies, and the recipe for the next low-resource languages.
• ABA Technology (Morocco): Built a Fusion AI Twin that sees, speaks, and predicts, combining live sensors, Omniverse 3D simulation, NVIDIA Cosmos 3 vision Reasoner, Isaac Sim, and Nemotron 3.5 Lightning to intelligently manage assets across the entire industrial chain. The digital twin, powered by the NVIDIA NeMo Agent Toolkit, becomes a central operational brain that anticipates process instability, coordinates decisions, and balances production, stability, and availability.
• Disseqt AI (Ireland): Built an on-premises AI compliance platform for the public sector, running entirely on sovereign infrastructure and powered by NVIDIA Nemotron 3 Super, Nemotron 3 Content Safety, and NeMo Guardrails. Every AI decision is checked, explained, and fully sealed as regulator-ready (EU AI Act–ready) evidence. Validated on the Irish Social Welfare Assistant use case.
• Algoritmica GmbH (Germany): Solved the overwhelming workload of loan applications for credit risk officers at German public banks, creating bottlenecks that delay capital deployment to small and medium enterprises. The team built an AI assistant platform using NVIDIA Nemotron 3.5 Lightning, Nemotron 3 Super, and Nemotron 3 Embed. The AI assistant writes credit memos for German public-bank underwriters and produces evidence the bank and its supervisor can recheck, slashing review times and accelerating lending. 

What the Codefest revealed

Across fourteen very different projects, common themes emerged. Data control is a product requirement, not an afterthought: for teams in financial services, healthcare, and regulated industries, keeping sensitive data within their own jurisdiction was non-negotiable. Open model weights made that feasible. Small, specialized models often outperform large generalist ones: targeted fine-tuning on the domain specific data,  and the hardest work is around the model, not inside it: data curation, agentic orchestration, evaluation harnesses, and compliance tooling were the areas where teams found their real gains leveraging NVIDIA AI software stack and NVIDIA Nemotron open models and libraries  for data curation, synthetic data generation, training, inference, and agent development.

Developing and testing AI models requires access to suitable computing infrastructure. During the Codefest, T-Systems supported participating teams with access to NVIDIA RTX PRO 6000 GPUs through its Industrial AI Cloud. Located in Munich and operated by Deutsche Telekom, Industrial AI Cloud provides high-performance NVIDIA GPU infrastructure for demanding AI workloads, including model training, inference, simulation and digital twins.

The Codefest offers a practical example of how access to advanced computing resources, combined with developer expertise and open AI technologies, can support innovation across different industries. 

What comes next