AI should be built for real use.

We believe the value of AI depends on how well its capabilities become part of everyday products. People need explanations they can understand, conversations they can practise and guidance they can act on.

Building those experiences takes choices about how an interaction feels, how it fits into a product and what it takes to keep it running. We see those choices as connected. The architecture behind an experience influences who can use it and whether its economics make sense as usage grows.

Start with the interaction people need.

We believe a visual presence can be useful when a task benefits from seeing who is speaking: practising a difficult conversation, exploring an explanation or finding a way through an unfamiliar service. Its value should be judged in the context of that task. Does the interaction help people understand, participate or make progress?

Make the architecture serve the experience.

This belief shapes MeshLite. It gives AI products a live visual presence through an embeddable avatar layer, with conversational logic and visual animation treated as separate parts.

When avatars are rendered continuously in the cloud, expensive GPUs keep working throughout live conversations. Longer sessions increase rendering work; more simultaneous conversations require more capacity. Those costs influence the experiences a product can afford to offer.

MeshLite starts by separating avatar creation from live use. We prepare an avatar once, then reuse it across separate conversations. Each user's browser brings the avatar to life in sync with speech, without continuous cloud GPU rendering.

This allows us to offer the visual layer without a per-minute rendering fee. A learner can make another attempt or a visitor can ask another question without adding another minute to that rendering bill. MeshLite subscriptions and plan limits still apply, and voice and AI services have their own costs.

Let research earn its place in the product.

We believe research is most valuable when it changes what people can actually use. For Toda Labs, that means investigating avatar motion, speech synchronization and browser performance together: how an avatar expresses itself, stays aligned with speech and works across devices.

A useful experiment makes a question clearer. It shows its conditions, reveals a trade-off and helps determine the next product decision. We aim to explain what we are investigating, show what the evidence supports and be precise about what remains unresolved.