Press Space to reveal
Explainable
AI System
Trust, but verify.
Every decision. Explained. Verified.
Why Explainability
is Critical
In critical and regulated environments, opacity is not acceptable.
X·AIS Paradigm
a native explainability layer
At Devana, AI systems produce results that are deeply inspectable, auditable over time, and never reconstructed after the fact.
This comes from an architectural choice:
explainability is not an add-on, nor a third-party tool, but a native layer built into the core of the system.
X·AIS Paradigm
Explainable AI Systems by Design
At Devana, AI systems produce results that are deeply inspectable, auditable over time, and never reconstructed after the fact.
This comes from an architectural choice:
explainability is not an add-on, nor a third-party tool, but a native layer built into the core of the system.
Transparency
in real time
Traceability
end to end
Reliability
hallucination detection
Compliance
continuous audit trail
Trust
measured and verifiable results
When AI becomes operational,
trust must become a system property.
This is why every layer of Devana is designed to make trust measurable, verifiable, and provable.
RAG Score™
Trust through measurement.
RAG Score™
Trust through measurement.
≥ 80%
50-79%
< 50%
RAG Score™ turns the trustworthiness of each response into a measurable and auditable property.
Every response is verified against what the system retrieves, interprets, and generates. Based on predefined trust thresholds, responses can be allowed, filtered, blocked, or escalated.
Genesis
Born from pioneering work in generative AI.
Protected Innovation
Grounded in French patent FR3136298 (INPI).
ODIN
Observational Document
Intelligence Network
An AI-native IDP technology that transforms unstructured documents into verified, structured data directly usable by AI systems.
INSURANCE CLAIM
Policy AUTO-2024-07832
Official Document
Policyholder Information
Damage Assessment
Front bumper, headlight, hood realignment. Third-party liability under assessment.
P. Moreau
Page 1/1
Structured Output
Beyond traditional IDP
Where traditional IDP solutions stop at OCR and extraction, ODIN interprets, structures, and understands, powered by its unique hybrid technology combining OCR rigor with Computer Vision power.
ODIN is the traceable document foundation of Devana's explainable AI systems.
Key Capabilities
ODIN transforms documents into reliable, actionable knowledge for AI systems.
Handles 20+ formats, including PPT, DOC, PDF, XLSX, STL, EML, HEIC, AI, video, and audio.
ODIN
Observational Document
Intelligence Network
An AI-native IDP technology that transforms unstructured documents into verified, structured data directly usable by AI systems.
INSURANCE CLAIM
Policy AUTO-2024-07832
Official Document
Policyholder Information
Damage Assessment
Beyond traditional IDP
Where traditional IDP solutions stop at OCR and extraction, ODIN interprets, structures, and understands, powered by its unique hybrid technology combining OCR rigor with Computer Vision power.
ODIN is the traceable document foundation of Devana's explainable AI systems.
Key Capabilities
ODIN transforms documents into reliable, actionable knowledge for AI systems.
Full Traceability Interface
Your results, ready to be audited. Nothing is hidden.
Every AI decision, broken down in real time.
From a single debug session to enterprise-wide deployment, Devana exposes every step of AI execution. LLM calls, retrieved sources, tool invocations: all visible, measurable, and auditable.
So that every AI decision is understood, verified, and justified.
See our governance framework and SLAsWhat You Can See
1,247
89%
3.2s
[INFO] LLM Call → mistral-large (prompt: 1,247 tokens)
[INFO] ✓ Vector search completed (189ms) — 8 sources found
[INFO] Tool Decision → extract_clauses(doc="contract_v3.pdf")
[INFO] ✓ RAG Score: 91% — Grade A
[INFO] ✓ Response generated (1.2s) — 892 tokens
1,247
3.2s
89%
2.4M
0.3%
[INFO] LLM Call → mistral-large (prompt: 1,247 tokens)
[INFO] ✓ Vector search completed (189ms) — 8 sources found
[INFO] Tool Decision → extract_clauses(doc="contract_v3.pdf")
[INFO] ✓ RAG Score: 91% — Grade A
[INFO] ✓ Response generated (1.2s) — 892 tokens
Activity by Hour
What this enables
Technical Debugging
Trace every error back to its source. Identify bottlenecks, step by step.
Regulatory Audits
Justify every decision and automatically generate the audit trail regulators expect.
Governance & Accountability
Track AI behavior, enforce your policies, ensure responsible deployment.
Continuous Supervision
Monitor your AI systems in real time. Catch issues before they reach your users.
Nothing is hidden. Everything is explained.
Full transparency at every layer of the AI stack
Trust is not declared.
It is built.
A trust metric must be continuously tested. It is never set in stone.
×
Devana continuously stress-tests and challenges its scoring systems, notably through an applied research program conducted with LIST Luxembourg.
Since 2025, Tom Lucas has been conducting a PhD thesis dedicated to designing and validating evaluation protocols for trusted information.
Tom Lucas
AI Researcher at Devana
"Explainable AI for Generative Models in RAG Contexts"
Devana × LIST Luxembourg
This research advances on four fronts:
Multi-metric evaluation
Beyond single-score assessments
Token-level analysis
Granular inspection of every generation step
LLM-as-judge approaches
Cross-evaluation by language models under transparent protocol
Transparent benchmarking
Reproducible and open evaluation protocols
Our goal is not to promote a score,
but to ensure its methodological validity.
From Explainability
to Sovereignty
Explainability in Devana's AI systems is the culmination of a long-term trajectory. Not a late response to regulatory pressure.
Foundations
Early RAG systems built for document analysis and fact-checking, before hallucinations and XAI became mainstream topics.
Patent & Trust Scoring
Filing of patent FR3136298 and introduction of the RAG Score™ to measure AI reliability instead of assuming it.
Real-World Stress Testing
Collaboration with investigative journalists to confront AI systems with adversarial information and complex real-world scenarios.
ODIN & Regulated Environments
Development of ODIN in collaboration with a telecom regulatory authority, pushing document intelligence toward compliance-grade robustness.
Scientific Validation
Academic partnership with LIST Luxembourg to validate explainable AI evaluation frameworks in RAG contexts.
Multi-Scale Deployment
Deployment of sovereign Sandbox environments with LuxProvide, and introduction of DIWY, a miniaturized sovereign AI making explainability accessible at every scale.
XAI becomes deployable at every scale:
from sovereign HPC to miniaturized AI.
Ready to inspect AI decisions,
as they happen?
Discover Devana's explainable AI systems.