With Light — DBT tools + guided AI coaching

Summary

With Light is a DBT-first emotional wellness product that helps users track emotions and situations, then guides them to the DBT skills and exercises they need in the moment.

The problem

When someone is dysregulated, they need help quickly. The product experience needs to be:

• fast (few taps to log)
• guided (clear next steps)
• structured (skills are evidence-based, not generic advice)
• safe (sensitive data, careful UX)

Constraints

• The system must support a growing library of DBT skills/exercises without hard-coding flows.
• “Personalization” needs to be explainable and controllable (avoid random-feeling recommendations).
• Privacy expectations are high (users are logging sensitive situations).

What I shipped

• A skills library that can power multiple surfaces (browse, search, “recommended for you”, guided exercise flows)
• A matching layer that maps user state (emotion + context + distress) → relevant skills and next steps
• Guided exercises that break down DBT tools into step-by-step interactions

Architecture (high level)

• A content model representing skills, exercises, and prompts as composable modules
• A recommendation/matching subsystem with deterministic fallbacks and guardrails
• Telemetry designed around product iteration (drop-offs, completion rates, time-to-first-tool)

Outcomes

• A product foundation that supports adding new DBT tools quickly and consistently
• Clear separation between content, matching logic, and UI surfaces (improves maintainability and experimentation)

What I’d do next

• Add lightweight experimentation (A/B) around tool matching and exercise completion
• Expand “insights” to help users see patterns while keeping the UX simple and non-judgmental