01
Unexplainable outputs
A generic LLM answer cannot be reconstructed. Under the transparency and oversight duties, “the model said so” is not a defensible record.
01EU AI Act · Regulation (EU) 2024/1689
EU AI Act Compliance Starts With Evidence, Not Explanations.
The Act asks you to show how an output was produced, who governs it and where the data lives. Deepentix addresses all three by design: CLAIM Graph technology and Vertical Small Language Models turn your regulated documentation into auditable Knowledge Assets with a 100% chain of evidence, deployable inside your own infrastructure.
ISO / IAIS
Solvency II Art. 132
Branch exception 4.2
Contract cap €80k
ResolvingL1
100%Chain of evidence
5 tiersHierarchical authority resolution
15–45×Lower cost
100%On-prem / sovereign ready
02The compliance gap
Why Generic LLMs and Vector RAG Cannot Carry the Burden of Proof
Transparency, record-keeping and human oversight are properties of an architecture, not a disclaimer added to a chatbot. Probabilistic retrieval produces answers nobody can reconstruct.
02
Similarity is not authority
Vector RAG retrieves what reads alike, not what legally governs. It silently answers from an EU-level framework when a national rule, corporate policy or contract override applies.
03
Data leaves your control
Third-party frontier APIs move regulated content outside your jurisdictional and contractual perimeter, colliding with GDPR and sector rules.
03Obligation mapping
Where CLAIM Graph and Vertical SLMs Meet the Act
A working map from the high-risk obligations of Regulation (EU) 2024/1689 to the platform capability that produces the evidence. Legal classification of your specific system remains your responsibility.
| Reference | Obligation | What the Act requires | Deepentix capability |
|---|---|---|---|
| Art. 10 | Data & governance | High-risk systems must be built on data that is relevant, representative and documented, with governance over its sources. | Ingestion produces a structured CLAIM Graph in which every extracted claim keeps its document, version and source tier—so the knowledge base itself is inventoried, not an opaque embedding blob. |
| Art. 11 · Annex IV | Technical documentation | Providers must maintain technical documentation describing the system's architecture, data and expected behaviour. | Ontologies (MVOs) and graph schemas are explicit, human-readable artefacts you own and can export as documentation evidence. |
| Art. 12 | Record-keeping & logging | Systems must technically allow the automatic recording of events over their lifetime to enable traceability. | Every answer resolves to a deterministic chain of evidence down to the line item, making each output reconstructable after the fact. |
| Art. 13 | Transparency to deployers | Operation must be sufficiently transparent for deployers to interpret the output and use it appropriately. | Outputs are delivered with citation provenance and the authority tier they were derived from, instead of unsourced generated prose. |
| Art. 14 | Human oversight | High-risk systems must be designed so that natural persons can effectively oversee and, where needed, override them. | Human-in-the-loop validation of ontologies at ingestion, and reviewable evidence at query time, keep domain experts in control of what the system may assert. |
| Art. 26 | Deployer obligations | Deployers must use high-risk systems per instructions, monitor operation and keep generated logs. | Knowledge Assets run inside your own infrastructure, so monitoring and log retention stay within your existing controls. |
Primary source: Regulation (EU) 2024/1689 on EUR-Lex.
04Chain of evidence
Every Answer Reconstructable, Line by Line
Deepentix resolves hierarchical authority at ingestion. At query time the system returns the governing source, its tier and the path taken to reach it.
100%
L6
Evidence, not assertions
Reasoning is resolved at ingestion into a CLAIM Graph, so every output carries a traceable path back to the governing source line.
Expert-validated ontologies
Minimum Viable Ontologies are bootstrapped automatically and confirmed by your own domain experts—oversight is designed in, not bolted on.
Sovereign by construction
Vertical Small Language Models run on-premises or in your private cloud with no dependency on third-party frontier APIs.
05Sovereignty & infrastructure
Regulated Data That Never Leaves Your Perimeter
Built from the ground up for strict European data compliance (GDPR, EU AI Act, Solvency II), with no reliance on third-party frontier APIs.
On-Premises & Private Cloud
Runs natively inside Microsoft Azure, Google Cloud, Databricks, AWS, or fully isolated air-gapped on-premise environments.
Non-Frontier Independence
Vertical Small Language Models remove dependency on third-party public cloud APIs, keeping sensitive corporate IP confidential.
Inherently Portable
Export and own your knowledge graphs and ontologies unconditionally. No vendor lock-in.
06Where this applies
Audit-Ready AI by Regulated Sector
Insurance & BFSI
Claims, underwriting and regulatory reporting where EU AI Act duties land alongside DORA and Solvency II.
Insurance & BFSIPharmaceuticals & Life Sciences
MLR validation and medical writing where citation accuracy is itself the compliance artefact.
Life SciencesSystems Integrators
Deliver audit-ready AI programmes for regulated clients without rebuilding governance per engagement.
Partner Program07Deployment starts here
Show Your Auditor the Path, Not the Prompt.
See auditable AI on your own regulated documents, with the full chain of evidence behind every answer.
- Obligation-by-obligation walkthrough on your own document set.
- Hierarchical authority resolution across EU, national, corporate and contract layers.
- Line-item chain of evidence for every answer.