Bantr
A mobile-first conversational game engine designed to help people skip small talk and reach better conversations in minutes.
Snapshot
A live production build tested under conversational conditions where rhythm and coherence matter.
Mobile-first SPA
React + Vite production deployment
React + Express + Postgres
TypeScript frontend with session-aware API/backend
Prompt generation
OpenAI-generated rounds constrained by format rules
Stripe subscriptions
Real payment path integrated into app flow
Guest-first identity
Optional accounts layered on top of instant play
Live environment
Shipped loop with tracked completion metrics
The Problem
Most conversations default to work, weather, politics, or logistics. Bantr was designed for moments where people want depth but do not know how to initiate it.
Scenario
Third date
People want depth but default to safe small talk.
Scenario
Hostel bar
Strangers need a fast path to meaningful interaction.
Scenario
First dinner with new friends
Conversation stalls without a shared creative prompt.
Why This Matters
As AI makes cognition cheaper, creativity becomes more valuable. Bantr was an experiment in building a product to help people practice creativity.
Goal
Better conversations through structured creativity.
Design Constraints
Human interaction has low tolerance for friction. Bantr trades open-ended question generation for a fixed 10-question structure so pacing and coherence stay stable.
Interaction Risks
- Latency breaks rhythm
- Unstructured AI output creates awkwardness
- Over-structuring removes spontaneity
- Payment gating must stay invisible
- State bugs immediately reduce trust
Production Guardrails
3-second response target
Question fetch with cached fallback
Moderation gate
OpenAI moderation before prompt display
Cookie session tracking
Progress and completion stability across refresh
AI tools accelerated implementation, but reliability depended on disciplined rule design and safeguards.
The System
Bantr runs on a round engine that remembers progress: fixed 10-question sessions, turn-by-turn continuity, and prompt formatting + validation.
Interaction Trigger
Low-friction social moment
No onboarding barrier
Prompt Pipeline
OpenAI prompt generation
Format cleanup + fallback cache
Safety Layer
Moderation pass
Prompt validation before render
Round Engine
Fixed 10-question flow
Turn-by-turn state continuity
Outcome Surface
Personalized summary
Completion tracking
Outcome and Future Direction
The product works in production, but adoption remains limited without a reliable acquisition path.
Core loop in production
- Users complete full 10-question rounds in production
- Personalized summaries generated at session completion
- Session and completion metrics logged for product learning
Distribution is separate work
The core loop shipped and worked, but adoption stayed limited without a defined distribution channel.
Distribution is a separate system with its own constraints. After product reliability, the next milestone is finding a reliable way to reach new users.
Future
Multimodal rounds
Expand from text-only prompts into image-building and travel-planning conversational formats.
Future
Stronger AI pipeline
Improve question generation and scoring so prompts feel less generic and more context-aware.
Future
Creator infrastructure
Create a way for others to launch new conversational games on top of the same system.