Eolas

15 AI coaches for career, fitness, wellness, relationships & personal growth. They remember you, track your goals, and work together.

Comment

Hackathon Application

About the Project

Inspiration

The coaching industry is a $20B+ market, yet access to quality coaching remains limited by cost - a single session with a professional coach can run $150-$500/hour. Most people don't just need help in one area; they need guidance across career, fitness, mental wellness, finances, and relationships simultaneously. We asked: what if everyone could have their own personal board of advisors?

The name Eolas (pronounced OH-lass) comes from the Irish word for knowledge and understanding. It reflects the app's core mission - not just providing information, but fostering genuine understanding across every area of a user's life.

The spark came from a simple frustration: generic AI assistants know a little about everything but lack the depth, personality, and continuity of a real coaching relationship. We wanted to build something where each coach remembers you, has a distinct voice and personality, and works within proven coaching frameworks - not just pattern-matched responses.

What It Does

Eolas is a mobile AI coaching platform (iOS/Android) that gives users access to a team of 15 specialised AI coaches, each with:

  • Distinct expertise - from career transitions (GROW Model) to sleep optimisation (Circadian Rhythm Science) to entrepreneurship (Lean Startup)
  • Unique personality and voice - ultra-realistic voices powered by ElevenLabs, so each coach sounds as different as they think
  • Separate memory - Maya remembers your job interview prep; Riley tracks your fitness journey; Casey knows your stress triggers. They maintain independent relationships with the user
  • Cross-coach intelligence - a weekly AI synthesis that connects the dots: "Your work stress (Casey) may be affecting your sleep (Morgan) and workout consistency (Riley)"

The subscription model offers 3 free coaches with 10 messages/day, while Premium unlocks all 15 coaches, unlimited messaging, ElevenLabs voices, and cross-coach insights.

How We Built It

The entire application - ~28,000 lines of TypeScript across 132 source files - was built in roughly 3 weeks (99 commits from January 17 to February 8, 2026).

Client: React Native 0.81 + Expo SDK 54 with file-based routing (Expo Router 6). State management via Zustand. Animations with React Native Reanimated 4.1.

Backend: Fully serverless on Firebase - Authentication (email, Google, Apple Sign-In), Firestore for real-time data sync, Cloud Functions for secure AI/voice proxying, Cloud Storage for audio caching, and Cloud Messaging for push notifications.

AI Layer: Anthropic's Claude API (claude-sonnet-4) powers all coaching conversations. Each of the 15 coaches has a bespoke system prompt encoding their personality, expertise, coaching frameworks (e.g., CBT techniques for Casey, Radical Candor for Jordan), and safety guardrails including crisis detection and professional referral protocols.

Voice: ElevenLabs for text-to-speech (each coach has a unique voice ID), OpenAI Whisper for speech-to-text. Includes a full hands-free conversation mode with waveform visualisation.

Payments: RevenueCat for cross-platform subscription management with webhook-driven entitlement sync.

Design System: A "Bold Editorial" aesthetic - high-contrast black and paper white, signal orange accents, sharp geometry with zero border-radius, and a three-font typographic system (Playfair Display for headlines, Outfit for body, Space Mono for labels). Every coach has a signature colour identity.

Challenges We Faced

1. Multi-coach context isolation. The hardest architectural decision was how to manage separate memory per coach while still enabling cross-coach insights. Each conversation maintains its own history with periodic summarisation (compressing after 50 messages), and the weekly synthesis is generated by a scheduled Cloud Function that carefully aggregates across coach contexts without leaking private details between them.

2. Voice at scale without burning money. ElevenLabs voices sound incredible but cost real money per character. We implemented aggressive audio caching in Firebase Cloud Storage and designed the architecture so common responses could be cached and reused, targeting a $50\%+$ cache hit rate.

3. Safety-first AI coaching. AI coaches operating in sensitive domains (mental wellness, relationships, parenting) demanded serious safety guardrails. Every coach includes crisis detection - if a user mentions self-harm or suicidal ideation, the coach acknowledges feelings with empathy, encourages professional support, and surfaces localised crisis resources. Coaches are explicitly instructed to never provide methods, never promise confidentiality for safety concerns, and to refer to licensed professionals for clinical issues.

4 Real-time UX polish under a hackathon deadline. Optimistic message updates, typing indicators, voice recording with live waveforms, onboarding animations, confetti celebrations on goal completion - making a $\$25$/month app feel premium in 3 weeks meant ruthless prioritisation and a component library that earned its keep.

What We Learned

  • System prompts are the product. The difference between a generic chatbot and a compelling coach is entirely in prompt engineering - personality, frameworks, tone, boundaries, and safety rails. We iterated on these more than any UI component.
  • Serverless is a hackathon superpower. Firebase + Cloud Functions let a small team ship a production-grade backend without touching infrastructure. The trade-off is vendor lock-in, but for a 4-week sprint, the velocity was worth it.
  • Voice changes everything. The moment we heard the first ElevenLabs-powered coach response, the app went from "another AI chat" to something that felt genuinely personal. Voice is underrated as a differentiator.

What's Next for Eolas

Short-term (1–3 months):

  • Accountability coach - A dedicated meta-coach that works across all goals and targets, helping users prioritise, avoid overcommitting, and maintain sustainable progress
  • Journalling — Free-form journal entries that give coaches out-of-chat context, enabling richer conversations and better progress tracking over time
  • Wearable integration - Apple Watch, Fitbit, and Oura Ring data to inform coaching context - sleep data for Morgan, activity data for Riley, stress signals for Casey
  • Streak mechanics and coach relationship milestones - Visual streak tracking and surfacing milestones like "You've had 50 conversations with Maya"
  • Offline support - Queue messages when offline, sync when reconnected

Medium-term (3–6 months):

  • Multi-modal input - Photo sharing for fitness check-ins, food logging, and document review
  • Coach collaboration - Two coaches participating in a single conversation for complex cross-domain topics
  • Coach marketplace - Community-created coach templates with revenue sharing
  • Adaptive coaching style - Coaches adjust communication style based on observed user preferences
  • Widget support - iOS/Android home screen widgets showing daily coach tip or streak

Long-term (6–12 months):

  • Enterprise tier - Company-sponsored coaching for employee wellness programs
  • API platform - Let third-party apps integrate Eolas coaching into their products
  • Predictive coaching - Anticipate user needs before they arise based on behavioural patterns
  • Research partnerships - Collaborate with universities to validate coaching methodology effectiveness

Built With

Share this project:

Updates