Overview
Creatrix is a professional creative agent that understands user patterns, strengthens relationships, and generates personalized coaching action items through questionnaire-driven flows and structured progression phases.
-
Problem
Scale personalized coaching without losing consistency
-
Stage
Prototype with active user onboarding
-
Stack
React · Python · Claude · prompt templates
Demo
Constraints
- Multi-turn coaching flows must stay coherent across phases and sessions.
- Prompt behavior needs to be editable by non-engineers through an admin UI.
- Output quality must be measurable, not assumed — feedback loops are part of the product.
- Prototype stage: ship learning loops while onboarding individual users.
Architecture
Daily coach output follows a structured pipeline from user inputs through context assembly to generated timeline items.
Prompt system
- Editable phase prompt templates in an admin panel.
- Context management aligned with the LUCID framework and multi-turn AI patterns.
- Validation layers to enforce consistency across coaching turns.
Feedback loop
- Daily questionnaire responses with thumbs up/down scoring and comments.
- Signals used to refine prompt templates and coaching effectiveness over time.
Key decisions
Templates over one-off prompts
Prompts are versioned, editable assets — not hard-coded strings — so product and coaching stakeholders can iterate without redeploying application code.
Validation before generation
Context assembly and guardrails run before model calls to reduce drift in multi-turn flows and keep phase progression predictable.
Measure coaching, not just usage
Explicit feedback on daily outputs creates a closed loop for prompt refinement instead of relying on anecdotal quality checks.
Screenshots
Outcome
- End-to-end platform in prototype with individual user onboarding underway.
- Admin tooling for prompt iteration without engineering bottlenecks.
- Structured pipeline from questionnaire inputs to actionable coaching items.