AI productIn development

Pravia

A bilingual AI product for practising emotionally difficult situations through adaptive text and voice simulations.

Role
Product definition, UX/UI, AI architecture, full-stack engineering and safety design
Engagement
In development
Pravia adaptive emotional training product homepage

The central challenge

Create realistic AI practice that feels engaging and useful while remaining explicitly safe, bounded and non-therapeutic.

Constraints

  • Streaming adaptive conversations in English and Portuguese
  • Safety classification, crisis response and intensity controls
  • Voice options with browser and cloud-provider fallbacks
  • Subscriptions, usage limits and configurable product operations

Scope

  • Product strategy
  • AI experience design
  • Full-stack development
  • Safety architecture
  • Subscription systems

The challenge

Most emotional-skills products explain what a person should do. Pravia is built around practice. It places users inside realistic situations—rejection, uncertainty, criticism, conflict or setting a boundary—and adapts the scenario in response to their choices.

That creates a more useful product, but also a more demanding one. The system needs convincing AI behaviour, calm interaction design and safety controls that remain effective across two languages and multiple model providers.

Safety is part of the architecture

Pravia is clearly positioned as training, not therapy. Stop and pause controls remain available throughout a session. Keyword screening and optional model classification can detect crisis language, pause the experience and surface localised help resources.

Intensity limits, de-escalation instructions and incident review are designed into the product model rather than added as policy copy after the fact.

The system behind the experience

The user experience includes onboarding, scenario selection, streamed simulation, voice interaction, debriefs, progress and achievements. Behind it sits a configurable administration system for prompts, capabilities, plans, languages, users, safety resources and provider settings.

Stripe subscriptions and usage gating support a freemium model. Browser speech keeps voice accessible without extra cost, while higher-quality providers can be enabled selectively.

Designed for practice

The visual system avoids gamifying vulnerability. It uses clear pacing, deliberate language and visible control to make difficult practice feel contained. The product is built around user needs rather than AI novelty.

Need help with a complex system?

Tell us what the product or system needs to achieve, where the main risks are, and what has made it difficult.