Amarena Solutions
Cardiac & Thoracic Surgery · HDS AI Factory
Feb 17, 2026
Voices in this story




Models used
Qwen 3 · Self-hosted
Conversational AI for clinical teams
MedGemma · Self-hosted
Medical document processing & analysis
Bio Mistral · Self-hosted
French sovereign model · Medical research
Technology partners
01 · Healthcare · AI in the Operating Room
Amarena Solutions partnered with Iguana Solutions to deploy a sovereign, HDS-certified AI Factory — built on Dell Technologies servers, Nvidia GPUs and the IG1 FullStack AI platform. The mission: use data intelligence to serve clinical care without ever compromising patient data sovereignty.
2
Clinical use cases
deployed
3
Specialised
AI models
Certified &
GDPR compliant
Two complementary use cases
AI models deployed
"We set up an Open Source LLM model with Iguana Solutions to be able to process surgical reports. The results we obtained were very satisfactory. The priority remains: using medical data in a sovereign and compliant framework, in close collaboration with hospital teams."
"We supported Amarena in setting up an AI Factory in an HDS and sovereign environment. Beyond all the technical aspects, Amarena relied on our expertise to choose the right models for each specific clinical use case."
EasyBourse
Online Brokerage · La Banque Postale
Feb 17, 2026
Voices in this story


Technology partner
02 · Finance · Automated Document Reading
EasyBourse — La Banque Postale's high-availability online brokerage — partnered with Iguana Solutions to tackle a massive KYC challenge: automatically categorising, reading and structuring over 1.2 million pages of client documents. The result is a multi-stage AI pipeline that handles 30+ document types with over 98% accuracy.
1.2M
Pages
processed
98%+
Document
classification accuracy
30+
Document
types supported
+40pts
MRZ zone
accuracy gain
The AI pipeline
"The main need was to reprocess 1.2 million pages of client documents — categorising them, analysing their content, structuring the extracted information and producing exploitable statistics. The model now detects document types with over 98% accuracy."
"The MRZ zone — designed for machines, paradoxically harder for language models to read. The optimisations on image size and rescaling brought a performance gain of around 40 points. Fine-tuning is ongoing to reach 98%+ on this zone as well."
EasyBourse
Online Brokerage · La Banque Postale · 2025
Sep 16, 2025
Voices in this story

Technology partner
03 · Finance · Sovereign AI for Developers & Compliance
In 2025, EasyBourse tackled two major challenges: processing one million pages of KYC client documents to meet regulatory obligations, and embedding a sovereign AI assistant directly into developers' daily workflows — all within a fully on-premise, GDPR-compliant environment. No data leaves the company.
1M
Pages
back-documented
96
FTEs avoided
in 1 month
<2%
Error rate requiring
human review
Two sovereign AI use cases
"The equivalent of 96 full-time employees, done in one month. The error rate remained extremely low — less than 2% of cases required human verification. As for the developers' AI assistant, now named Albert — it's fully adopted by the teams."
"I found at Iguana Solutions a true collaboration — not a simple ticket-based process. The exchanges are direct, regular and constructive, on both infrastructure and AI topics. Teams deliver fast — which is rarely the case with other providers I've worked with."
Next ambitions
Paymium
Bitcoin Exchange · Founded 2011 · France
Feb 17, 2026
Voices in this story


Model used
Qwen-3 · Self-hosted
Deployed & managed by Iguana Solutions · Multilingual
04 · Fintech · Agentic KYC Automation
Paymium — France's leading Bitcoin exchange since 2011 — needed to automate a critical yet time-consuming KYC task: verifying that incoming IBAN transfers matched the registered client name. A process that took 3 minutes per case and only ran during business hours is now handled by an AI agent in 1 second, around the clock.
99%+
Accuracy on
1,000 accounts
1 sec
vs 3 minutes
previously
24/7
Always-on
availability
<5
Ambiguous cases
per 1,000
How the agent works
"Today, it does it in 1 second — and on top of that, it does it 24 hours a day, 7 days a week. Our platform is open 24/7; our teams are not. This gives us a response that is much faster and always running. Out of 1,000 accounts tested, fewer than five cases were genuinely ambiguous."
Next on the roadmap
Ready to build yours?
Whether it's document intelligence, autonomous agents, or sovereign AI infrastructure — our team of AI engineers is ready to scope your project and deliver measurable outcomes.