Lyon

Foundation models trained from scratch on your customer event data. Runs in your cloud. You own it.

[Lyon @lyondotso]

$LYON is live now on @ponsdotfamily

CA: 0x86ddf35636b277bff008725a767af8db57903c5d

Foundation models trained from scratch on your customer event data. Runs in your cloud. You own it.

https://t.co/Scs4xKfHLS

[Lyon @lyondotso]

AI should work with your data — without taking your data away.

Lyon is built to run inside your own cloud environment:

• Your data stays within your infrastructure
• Training and inference run inside your VPC
• Designed for enterprise security requirements
• Supports AWS, GCP, and Azure
• Your proprietary data remains yours

Lyon brings foundation-model intelligence to customer data while keeping control where it belongs.

Your data. Your cloud. Your intelligence.

Explore → https://t.co/chlz0hAIai

[Lyon @lyondotso]

What can Lyon power?

A single foundation model can support multiple areas of customer intelligence:

• Churn prediction
• Credit risk assessment
• Fraud detection
• Collections optimization
• Income prediction
• Cross-sell opportunities

Instead of creating a separate ML system for every problem, Lyon provides one intelligence layer that can support them all.

Customer data → intelligence → better decisions.

Learn more: https://t.co/chlz0hAIai

[Lyon @lyondotso]

Why is Lyon different?

• A foundation model built for customer event data
• Learns patterns across transactions and interactions
• Creates reusable customer embeddings
• Supports multiple prediction tasks from one model
• Deploys securely inside your own cloud

Instead of building ML systems from scratch for every problem, Lyon creates a reusable intelligence layer for the entire organization.

Explore Lyon → https://t.co/chlz0hAIai

[Lyon @lyondotso]

Traditional ML often starts with the same challenge:

• Build features manually
• Create a separate pipeline for every use case
• Retrain models as customer behavior changes
• Maintain multiple systems across the organization

This makes customer intelligence expensive, fragmented, and difficult to scale.

Lyon takes a different approach.

One foundation model.
One intelligence layer.
Multiple predictions.

That’s the shift Lyon is building.

https://t.co/chlz0hAIai

[Lyon @lyondotso]

How Lyon works.

• Connect your customer event data
• Train a foundation model on behavioral patterns
• Transform raw events into rich customer embeddings
• Fine-tune the model for specific business objectives
• Deploy predictions for churn, fraud, credit risk, collections, and cross-sell
• Run training and inference securely within your own cloud

The key: Lyon turns your proprietary data into a reusable intelligence layer — without moving your data outside your environment.

Your data. Your cloud. Your intelligence.

https://t.co/chlz0hAIai

[Lyon @lyondotso]

Have features that already work? Keep them, and upgrade them.

Lyon's downstream models can take your existing features alongside the embeddings.

[Lyon @lyondotso]

Your data never leaves your cloud.

Training and inference run inside your VPC, on AWS, GCP, or Azure. Complete data isolation. SOC 2, GDPR and LGPD.

Your data stays yours, and so does every model trained on it.

[Lyon @lyondotso]

BURNED ALL SUPPLY

https://t.co/War5PQmeNe

[Lyon @lyondotso]

$LYON is live now on @ponsdotfamily

CA: 0x86ddf35636b277bff008725a767af8db57903c5d

Foundation models trained from scratch on your customer event data. Runs in your cloud. You own it.

https://t.co/Scs4xKfHLS

[Lyon @lyondotso]

A new prediction shouldn't mean a new months-long project.

With Lyon you select a label table, verify the split, and start a LoRA fine-tune. The task adapter trains on top of the foundation model you already have.

[Lyon @lyondotso]

Churn, credit risk, fraud, collections, income, cross-sell.

One model. Not six feature engineering projects.

[Lyon @lyondotso]

Every customer leaves a trail of events: transactions, logins, card declines, support chats.

Lyon trains a foundation model from scratch on that event data. Events and profile go in. Customer embeddings come out.

Those embeddings are the starting point for every prediction that follows, and because the model runs in your cloud, the intelligence it learns belongs to you, not to a vendor.

That's the whole idea behind "foundation models on event data": train once on what your customers actually do, then build on top of it.

[Lyon @lyondotso]

Introducing Lyon.

Foundation models trained from scratch on your customer event data. The model runs in your cloud, and you own the intelligence.