New: our AI Act guide for SMEs is now available Download โ†’
VULYX VULYX
Free AI diagnostic โ†’
VULYX LAB ยท APPLIED R&D

We test tomorrow's AI, so you don't have to.

A space for applied research where we evaluate the latest models, invent new use cases, and publish our results โ€” successes as well as failures.

PUBLISHED EXPERIMENTS
06
RESEARCH AXES
05
IN PROGRESS
01
CADENCE
1 / month
WHY A LAB?

AI evolves every week and no SME has time to test everything. We do it for you, with rigor and transparency: protocol, data, results, limitations.

RESEARCH AXES
AXIS 0101 EXP.
Autonomous agents & orchestration
AXIS 0201 EXP.
Reliability and evaluation of language models
AXIS 0301 EXP.
Frugal AI: small models, reduced cost and footprint
AXIS 0401 EXP.
Vision & multimodality for industry
AXIS 0502 EXP.
Trustworthy AI: explainability and AI Act compliance

All experiments

6 RESULT(S)
TYPE
STATUS
EXP-014 Completed
Can an agent close a level-1 support ticket on its own?
KEY RESULT[72%] of tickets resolved without error across 400 cases
Autonomous agentsBenchmarkSep 20268 min
EXP-013 Completed
Small models vs. large models: where's the real sweet spot?
KEY RESULT[รท8] on inference cost for [-4 pts] of quality
Frugal AIBenchmarkAug 202611 min
EXP-012 In progress
Detecting a surface defect with 50 training images
KEY RESULTInterim results: [91%] recall
Vision & multimodalityPrototypeJul 20266 min
EXP-011 Completed
Measuring hallucinations in a document assistant
KEY RESULT150-question evaluation protocol published
LLM reliabilityResearch noteJun 20269 min
EXP-010 Completed
Explaining a forecast to a store manager
KEY RESULT[4 out of 5] users understand the forecast unaided
Trustworthy AIPrototypeMay 20267 min
EXP-009 Completed
An AI Act classification grid usable in 20 minutes
KEY RESULTGrid tested on [12] real projects
Trustworthy AIResearch noteApr 20265 min

We share

OPEN SOURCE ยท GITHUB
vulyx/llm-eval-kit[โ˜… โ€”] vulyx/ai-act-grid[โ˜… โ€”] vulyx/frugal-bench[โ˜… โ€”]
TALKS & PUBLICATIONS
[Talk title]
[Event ยท date]
[Publication title]
[Journal / conference ยท date]
[Talk title]
[Event ยท date]
COLLABORATE WITH THE LAB

Co-experiment with us

Companies, schools, researchers: propose a challenge, we'll build the protocol together.

Propose a challenge โ†’