Prept

• 

  logo

Inspiration

I wanted to create something that actually adapts to people’s real needs — whether you're preparing for a high-stakes interview, an unexpected emergency, or even just tracking your health. The idea was simple: instead of rigid templates, what if AI could understand your situation and build the tools you need?

That’s how Prept was born — an adaptive, intelligent preparation assistant.

What it does

You give Prept a simple natural language prompt (like “I have a big exam next week”), and it instantly generates:

• A personalized workspace tailored to your task, emotional state, and urgency
• Modular tools like smart checklists, study plans, progress trackers, calendars, and note spaces
• A themed UI with appropriate color schemes, icons, and tone
• Smart suggestions and timelines based on what you’re trying to achieve

Each section is dynamic and created from scratch — no predefined templates.

What I successfully built

• Dynamic section generation based on intent (exam, travel, interview, etc.)
• Theming engine that visually adapts to the user’s goal and emotional state
• Smart tool composer that selects appropriate tools (trackers, notes, reminders, etc.)
• Intent and emotion detection to adjust urgency and planning logic
• Quota store system to track usage of ElevenLabs and Tavus APIs, aligned with their free usage tiers

What I couldn’t build due to Bolt’s limitations

Some key features couldn’t be implemented because of technical limitations of the Bolt platform, not because of time constraints. These include:

• ❌ AI chat assistant: The core AI assistant couldn’t respond to messages due to Bolt’s lack of full context-passing and persistent chat history
• ❌ Voice narration with ElevenLabs: Although the quota system was implemented, actual playback and streaming of audio weren’t possible
• ❌ Video coaching with Tavus avatars: API integration wasn’t possible due to Bolt’s lack of media rendering and video upload support
• ❌ Persistent session memory: Dynamic context across user-created sections wasn’t supported, which limited the assistant’s ability to retain information
• ❌ Quotas UI: The quota system worked in logic, but Bolt's UI didn't render updates reliably in real-time

What I learned

• Bolt is a powerful tool for building UI-driven logic fast — but is not yet ready for apps that require conversational AI, dynamic memory, or real-time media rendering
• Trying to force high-context apps into no-code tools introduces more time cost than it saves
• Thinking modularly and preparing for graceful fallbacks is essential

What’s next

• Migrate Prept to a custom-built stack (likely Next.js + Supabase or Firebase)
• Fully enable voice and video coaching with ElevenLabs and Tavus
• Add persistent user memory and multiple dynamic sections
• Launch Prept as a powerful AI prep tool for anyone, anywhere

Built With

• bolt.new-(no-code-ui-builder)
• elevenlabs-api-(voice-synthesis)
• react-(custom-code-integration)
• tavus-api-(ai-video)
• typescript
• zustand-(state-management)