Hermes AI - AI Coach Marketplace

Logo

Inspiration

We chose Simon’s brief because the “minimalist, action-first coaching” direction felt both practical and personally meaningful. Most coaching apps overwhelm users with long responses. We wanted Hermes AI to deliver one clear next action in under a minute, then keep momentum through visible progress and gamification.

What it does

Hermes AI is a gamified AI coach marketplace that gives users instant access to specialized coaching for any goal. Users can:

Browse & use community-created AI coaches optimized for specific needs (productivity, wellness, focus, creativity)
Create & publish their own AI coaches with custom personas, tones, and coaching styles
Chat instantly with AI-powered coaches using Gemini and GPT fallback for fast, personalized guidance
Memory System: Chat memory with user approval flow to maintain personalized context across sessions
Build momentum through daily streaks, XP progression, and leveling systems
Complete weekly quests that reinforce positive behaviours and reward consistency
Compete on leaderboards (daily, weekly, monthly) to stay motivated alongside a community

It also includes a Simon-fidelity Decision & Strategy flow:

choose approach (conversational / structured / quick report),
complete mandatory scoping questions,
select a decision framework,
receive structured outputs (quote, synthesis, recommendations, immediate actions),
save decision reports in a Decision Log.

Every coach conversation is action-focused, designed to give you one immediate next step rather than overwhelming you with advice.

How we built it

We built Hermes AI as an iOS-first Flutter application with a clean architecture approach:

Frontend: Flutter with Provider for state management, custom UI components for a premium feel
Backend: Firebase (Firestore for data, Firebase Auth for anonymous-first onboarding)
AI Integration: Gemini API as primary LLM with GPT fallback via Wiro for reliability
Monetisation: RevenueCat integration with 2-tier paywall strategy (post-value paywall after first coaching session). There is a free tier with limited features like a daily message cap, less smart LLM and users can't create and share custom coaches. When the user subscribe they can use every feature without limits.
Gamification: Custom-built streak tracking, XP/leveling system, and quest progression engine

The architecture follows domain-driven design with clear separation between presentation, domain, and data layers, making it maintainable and testable.

Challenges we ran into

Achieving sub-6s response times while maintaining conversation quality required careful API optimisation and implementing smart fallback strategies between Gemini and GPT
Balancing user control over what the AI remembers vs. friction in the chat flow, we solved this with a non-blocking approval system that doesn't interrupt conversations
Designing quest difficulty and XP rewards that feel meaningful without being exploitable or overwhelming
UI overflow/performance tuning while keeping onboarding and chat animations smooth.
Creating a 3-step onboarding that delivers value in under 2 minutes while still capturing necessary user context

Accomplishments that we're proud of

Speed to value: Users reach their first AI coach response in under 2 minutes from app open
Polish: Built a premium glassmorphism UI with organic backgrounds and smooth animations that feels native iOS
Complete feature set: Shipped an end-to-end product loop: onboarding → first value → paywall → marketplace → chat → gamification.
Gamification that works: Quest completion paths are intuitive and players actually maintain streaks

What we learned

Essentialist design works: Stripping features down to "one deliberate action" creates higher engagement than feature-rich alternatives
Gamification needs balance: XP and streaks are powerful motivators, but only if rewards feel earned, not given
LLM outputs: Structured AI response schemas dramatically improve reliability and UX.
AI fallback strategies are critical: Having Gemini → GPT failover saved us from complete service disruption during API outages
Anonymous-first reduces friction: Letting users explore before signing up dramatically improved onboarding completion rates
Memory approval is a feature, not friction: Users appreciate control over what AI remembers about them
*Free tier *: For this kind of assistant apps, let the users try the app before buying anything