DreamChase

Turn your goals into structured action—with an AI mentor that ensures you actually achieve them.

Comment

Inspiration

In a world where information is abundant, execution is the real bottleneck.

Millions of learners set ambitious goals—becoming a competitive programmer, mastering AI, or cracking interviews—but fail due to lack of:

  • structured guidance
  • accountability
  • feedback loops

Existing tools like Notion or Trello help track tasks, but they do not ensure actual learning or progression.

This led us to ask:

What if AI could act not just as a guide, but as a mentor that ensures real progress?

What it does

DreamChase is an AI-powered goal execution engine that transforms ambitions into structured, adaptive learning journeys.

Core Learning Loop

$$ \text{Goal} \rightarrow \text{Tasks} \rightarrow \text{Execution} \rightarrow \text{Verification} \rightarrow \text{Progression} $$

Key Features

AI Roadmap Engine

  • Generates structured, level-based learning paths
  • Uses Google Search grounding to ensure reliable resources
  • Eliminates broken or hallucinated links

Multimodal Curriculum Extraction

  • Upload syllabus or textbook images
  • AI extracts and structures content instantly
  • Personalizes roadmap based on real curriculum

Live AI Mentor Verification

  • Powered by Gemini Vision
  • AI analyzes user’s screen/camera feed
  • Conducts real-time validation + questioning

Not just “Did you complete?”
But “Did you understand?”

Gamified RPG Progression

  • XP system, Gold Stars, and level unlocking
  • Progress tied to verified learning, not clicks
  • Immersive feedback loop to sustain engagement

How we built it

AI Core

  • Google Gemini 2.5 Flash for low-latency multimodal inference
  • Structured prompt pipelines for:
    • roadmap generation
    • evaluation logic
    • grounded search integration

Vision & Real-Time Interaction

  • WebSockets (Flask-Sock) stream live frames
  • Gemini Vision processes visual input in real time
  • Enables contextual, dynamic feedback

Voice Interaction Layer

  • Custom Text-to-Speech pipeline
  • Synced with UI animations
  • Creates a conversational mentor experience

Frontend

  • High-performance Vanilla JavaScript
  • Glassmorphism + neo-brutalist UI system
  • Real-time state synchronization

Backend

  • Flask-based architecture
  • Handles:
    • multimodal pipelines
    • AI orchestration
    • authentication via Firebase

Challenges we ran into

Real-Time Stream Consistency

Maintaining conversational continuity while streaming video frames introduced synchronization issues.

Solution:
Optimized frame buffering + event-driven state handling.

Link Hallucination Problem

AI-generated roadmaps initially included unreliable resources.

Solution:
Integrated Google Search grounding to ensure factual, working links.

Voice-UI Synchronization

Aligning speech output with UI animations across browsers was non-trivial.

Solution:
Fine-tuned animation listeners + audio event hooks.

Accomplishments that we’re proud of

Visual Verification Breakthrough

The AI could interpret real user work and respond contextually:

“I see your NumPy array—how did you reshape it?”

This transformed AI from a generator into a true evaluator.

Product-Level Experience

Built an interface that feels like an RPG—not a traditional learning tool.

Zero-Friction Onboarding

Users can:

  • Input a goal
  • Receive a roadmap
  • Complete and verify a task

All within 60 seconds

What we learned

  • Learning systems fail without feedback loops
  • Multimodal AI drastically improves engagement
  • Structured prompt engineering is critical for reliability
  • Low latency is essential for real-time AI interaction

What’s next for DreamChase

  • Collaborative verification (peer + AI hybrid)
  • Dynamic skill trees replacing linear roadmaps
  • Mobile-based real-world task verification
  • Deep skill analytics and performance tracking

Conclusion

$$ \text{Learning} = \text{Execution} + \text{Verification} $$

DreamChase bridges the gap between intention and achievement by ensuring users don’t just follow a path—but actually master it.

Built With

Share this project:

Updates