Evaluation Highlights

| Criterion | Threshold |

|-----------|-----------|

| Excellence | 4/5 |

| Impact | 4/5 |

| Quality & efficiency of implementation | 4/5 |

Cumulative minimum score: 12/15.

Key emphases:

* Scientific & technological excellence of GenAI approach.

* Credible pathway to clinical/industrial uptake.

* Robust ethics, gender & SSH integration.

* Consortium’s capacity to access multimodal data and HPC resources.

Strategy to Win the GenAI4EU Call

1. Craft a Mission-Driven Consortium

- Balance science, technology, clinical application and SSH.

- Mandatory profiles: AI/ML teams, clinical centres in two disease areas, data custodians, HPC provider, ethics & legal experts, SME/start-up developing GenAI tools, patient organisation.

- Demonstrate gender balance and pan-European coverage (at least one widening country partner recommended).

2. Anchor Objectives to Call Logic

Define 4–5 SMART objectives that map one-to-one to:

1. Development of new / repurposed GenAI models

2. Dual medical use-cases PoC

3. Continuous evaluation framework (performance, explainability, bias)

4. Researcher usability & training toolkit

5. Open-source release & policy contribution

3. Design a Coherent Work-plan (suggested WPs)

1. WP1 – Project, risk & quality management

2. WP2 – Data governance, FAIR compliance, GDPR & ELSI

3. WP3 – Multimodal data engineering & HPC pipeline

4. WP4 – GenAI model development & optimisation

5. WP5 – Use-case validation (two therapeutic areas, TRL 3-5)

6. WP6 – Continuous evaluation & alignment with human values

7. WP7 – Training, capacity-building, usability assessment

8. WP8 – Dissemination, exploitation, open science & clustering

4. Excellence Section Tips

- Provide state-of-the-art review & show clear novelty over existing GenAI frameworks.

- Detail data modalities, sample sizes, federated vs. centralised learning strategy.

- Explain bias-mitigation techniques, synthetic data generation, uncertainty quantification.

- Reference compliance with Ethics Guidelines for Trustworthy AI.

5. Maximising Impact

- Define KPIs: model AUROC, bias reduction %, compute efficiency, #researchers trained.

- Outline exploitation route: open-source licence, SME service packages, standardisation inputs (CEN-CENELEC, ISO), contribution to AI Act conformity assessment.

- Include sustainability plan: model upkeep after project, business models, integration with EHDS.

6. Robust Implementation & Resources

- Show resource allocation linked to tasks and realistic person-months.

- Allocate 5-7 % of budget for clustering meetings and joint activities.

- Include risk matrix with mitigation (data access, compute cost spikes, regulatory change).

7. Cross-cutting Priorities

- Integrate SSH experts in bias, societal acceptance, human-AI interaction.

- Apply gender dimension in datasets & analysis.

- Plan ethics self-assessment and external advisory board.

- Involve EC-JRC for regulatory science WP.

8. Budgeting Hints

- Compute & storage: justify HPC/GPU hours via AI Factory vouchers or commercial clouds.

- Personnel costs dominate (~60 %); include data stewards & DevOps staff.

- Dissemination: allocate open-access fees, conference booths, hackathons.

9. Writing & Submission Timeline

Milestone | Date

--- | ---

Consortium agreement draft | Jan 2025

Concept note freeze | Mar 2025

Full draft v1 peer-review | Jul 2025

Final internal review | Aug 2025

EU portal submission | Before 10 Sep 2025

10. Common Pitfalls to Avoid

- Under-estimating ethics & data privacy workload – allocate sufficient PMs.

- Missing explicit link between each deliverable and expected outcomes.

- Neglecting usability and researcher training – dedicate a full WP.

- Over-promising clinical deployment (keep within PoC, TRL ≤5).