Each contribution addresses a foundational barrier that prevents European regulated industries from operating sovereign, certified AI at scale. Each produces open-source artifacts with spillover across Member States and regulated sectors.
01
Open source
Federated Retrieval
Cross-border document retrieval across Member State regulatory corpora without exposing source documents. Regulatory intelligence without centralization.
What gets produced
A federated retrieval system combining hierarchical summarization with knowledge graph extraction. Each institution maintains its own corpus. The federated layer enables cross-border queries without either party exposing underlying documents.
02
Federated
Sovereign Mixture-of-Experts (SS-MoE)
Federated training across IPCEI partners. Each Member State trains on local data; the combined model serves any domain, any jurisdiction, with certified routing guarantees.
What gets produced
A mixture-of-experts architecture where expert modules are trained independently by different IPCEI partners on their local data. A sovereign router with conformal prediction guarantees directs each inference request to the appropriate expert without exposing training data across boundaries.
03
Infrastructure
Speculative Decoding for Sovereign Inference
2-3x inference speedup on sovereign European infrastructure, eliminating the latency penalty that drives enterprises toward hyperscalers.
What gets produced
A speculative decoding implementation optimized for the sovereign deployment constraint: limited GPU hardware, no cloud burst capacity, strict latency requirements from production workflows. Significant speedup without any change to output quality.
04
EU AI Act
Certified Uncertainty
Provable coverage guarantees for any AI model in any domain. Satisfying EU AI Act requirements with mathematical proof, not empirical benchmarks.
What gets produced
A conformal prediction toolkit that wraps any AI model with distribution-free uncertainty quantification. Output is not a point estimate but a prediction set with a guaranteed coverage rate. Provable bounds rather than empirical benchmarks.
05
Certifiable
Neurosymbolic Regulatory Reasoning
EU regulation encoded as machine-verifiable symbolic rules. AI systems that produce certifiable reasoning traces, not just answers.
What gets produced
A neurosymbolic architecture combining neural capabilities with formal symbolic reasoning. Regulatory text is encoded as machine-verifiable rules, not natural language prompts. Compliance determinations are auditable by design, not by assertion.
Compound Effects
Greater than the sum of parts.
The five contributions are designed to compound. Certain combinations produce capabilities that no single contribution delivers alone.
02 + 04
Model + Uncertainty
The sovereign model router uses certified uncertainty to guarantee selection accuracy. Each routing decision carries a coverage guarantee. The system knows when it is uncertain.
01 + 05
Retrieval + Reasoning
Federated retrieval feeds the neurosymbolic engine. Retrieved regulatory text is matched against formal symbolic rules, not passed as context. Two layers of auditability.
03 + 02
Acceleration + Architecture
Inference acceleration within each expert module. Distributed workload across specialized models. Combined: sovereign inference matching cloud-hosted alternatives on institutional hardware.
