HOPN Lab
Trustworthy AI under constraint.
How do we build AI systems that organizations can trust, audit and afford to run themselves?
HOPN Lab is the research agenda under the EHOPn platform.
HOPN Lab studies AI systems that stay reliable, private and verifiable under real-world constraints: limited compute, no cloud access, sensitive data, regulated domains and physical safety. Our work is organized as three layers (Route, Ground, Assure), one foundation program and one physical-world track.
This page describes our research agenda. Status labels show what is planned, in progress or complete.
Why this position
- 1. Frontier labs optimize capability at scale. Few groups own the question of verified deployment under constraint.
- 2. In Europe, GDPR, the EU AI Act and data sovereignty make constraint a requirement, not a weakness.
- 3. We publish our protocols and raw logs, and we disclose our commercial relationships.
Architecture stack
ConceptualFive blocks: Route, Ground and Assure stacked on Foundation, with Physical beside them.
The diagram is also listed as a table below.
View as table
| Program | Question | Direction |
|---|---|---|
| Route | How do we compose many models and providers behind one reliable interface? | Policy-driven routing across cost, latency, privacy and quality |
| Ground | What does the system know, and how do we prove where it came from? | Provenance-first knowledge graphs |
| Assure | How do we know each output is good enough, at runtime, before it ships? | Quality orchestration with contracts and calibrated methods |
| Foundation | What quality per watt and per euro is possible when data never leaves the organization? | Private and efficient AI |
| Physical | How do Route, Ground and Assure apply where errors have physical consequences? | Verified physical autonomy |
Research programs
- Planned
- In progress
- Complete
Track A
In progressRoute
How do we compose many models and providers behind one reliable interface?
Policy-driven routing across cost, latency, privacy and quality
Open programTrack B
In progressGround
What does the system know, and how do we prove where it came from?
Provenance-first knowledge graphs
Open programTrack C
PlannedAssure
How do we know each output is good enough, at runtime, before it ships?
Quality orchestration with contracts and calibrated methods
Open programTrack F
In progressFoundation
What quality per watt and per euro is possible when data never leaves the organization?
Private and efficient AI
Open programTrack D
PlannedPhysical
How do Route, Ground and Assure apply where errors have physical consequences?
Verified physical autonomy
Open program
Extract, Ground, Route, Assure, then Learn. Learning feeds back into Route and Assure.
End-to-end pipeline
ConceptualExtract, Ground, Route, Assure, Learn, with feedback from Learn to Route and Assure.
The diagram is also listed as a table below.
View as table
| Step | Feeds |
|---|---|
| Extract | Ground |
| Ground | Route |
| Route | Assure |
| Assure | Learn |
| Learn | Route, Assure |
Project to program matrix
Public projects mapped to programs. Private projects are omitted.
| Project | Route | Ground | Assure | Foundation | Physical |
|---|---|---|---|---|---|
| AI-PassInfrastructure layer for the other tracks | Mapped | Not mapped | Not mapped | Not mapped | Not mapped |
| TEAOATrust model for orchestration | Mapped | Not mapped | Not mapped | Not mapped | Not mapped |
| SLDE-AFT and offline LoRA extractionCorroborated extraction feeding the knowledge graph | Not mapped | Mapped | Not mapped | Mapped | Not mapped |
| SemvecTemporal and versioned memory | Not mapped | Mapped | Not mapped | Not mapped | Not mapped |
| ServAlignConstraints and policy as machine-checkable objects | Not mapped | Not mapped | Mapped | Not mapped | Not mapped |
| Invoice automationShared testbed | Mapped | Mapped | Mapped | Not mapped | Not mapped |
| SovraCompute-spectrum efficiency, private RAG, graph pruning | Not mapped | Mapped | Not mapped | Mapped | Not mapped |
The diagram is also listed as a table below.
View as table
| Project | Programs | Role |
|---|---|---|
| AI-Pass | route | Infrastructure layer for the other tracks |
| TEAOA | route | Trust model for orchestration |
| SLDE-AFT and offline LoRA extraction | ground, foundation | Corroborated extraction feeding the knowledge graph |
| Semvec | ground | Temporal and versioned memory |
| ServAlign | assure | Constraints and policy as machine-checkable objects |
| Invoice automation | route, ground, assure | Shared testbed |
| Sovra | foundation, ground | Compute-spectrum efficiency, private RAG, graph pruning |
Roadmap summary
Horizontal bars across 0-3, 3-6 and 6-12 months, grouped by program status.
AI-Pass submission
In progressQuality orchestration charter and shared invoice benchmark
PlannedSimulation baseline with fault injection (ROS 2, Gazebo)
PlannedGraph-pruning study
PlannedFirst quality-routing results
Planned
The diagram is also listed as a table below.
View as table
| Item | Program | Months | Status |
|---|---|---|---|
| AI-Pass submission | route | 0-2 | In progress |
| Quality orchestration charter and shared invoice benchmark | assure | 0-3 | Planned |
| Simulation baseline with fault injection (ROS 2, Gazebo) | physical | 0-3 | Planned |
| Graph-pruning study | ground | 3-6 | Planned |
| First quality-routing results | assure | 3-6 | Planned |
Evidence discipline
How the lab publishes: logs, protocols, disclosed relationships.
Governance
Governance wording is prepared and waits for director sign-off. It is not linked yet.
Work with HOPN Lab
Researchers can apply through EHOPn. The lab agenda is public. Results stay Pending until measurements exist.