VibeFlow AI Hackathon
Research Report

Solving EdTech's Personalization Gap: Connecting students with compatible mentors through AI-powered matching

Published: December 31, 2025

Where Human Intuition Meets AI Execution

Executive Summary

The Opportunity

The hackathon addressed two critical pain points: helping students find compatible, qualified mentors for niche subjects, and providing an efficient platform for expert mentors to share knowledge without administrative overhead.

The EduVibe Solution

An AI-powered EdTech platform with personalized student onboarding, AI-driven mentor recommendation with explainable AI (e.g., "90% match due to ML expertise"), interactive booking, and dual dashboards.

Core Finding

AI is indispensable for rapid prototyping and generating standard components. However, its use requires constant human validation, particularly for state management and complex decision logic.

Key Challenges Identified

State Loss

Data lost during form submissions, page refreshes, and navigation between steps

Logical Errors

Carousel active slide tracking failed; AI-generated logic insufficient for dynamic components

Over-Engineering

Unnecessary form steps and bloated state logic requiring manual cleanup

Outdated Library Suggestions

AI suggested deprecated libraries (moment.js) requiring replacement (day.js)

Complex Visual Interpretation

Animated graphics became simple placeholders; struggles with creative UI designs

Agent Mode Hardcoding

Custom logic via Agent Mode resulted in hardcoded values and inflexible implementations

Slower Response for Large Context

Claude 3 Opus showed slower response times when processing larger context windows

Domain-Availability Mismatches

GPT-4o occasionally mismatched domain expertise with mentor availability in recommendations

AI Models & Primary Tasks

GPT-4o

Mentor Matching (Task 3)

High accuracy in generating explainable AI output

Claude 3 Opus

Homepage & Dashboard (Tasks 1, 5)

Natural UI structuring with verbose explanations

V0 by Vercel

UI Generation & Preview

Rapid scaffolding with limited customization

Full Tool Stack Explored

Claude 3 Opus
GPT-4o
V0 by Vercel
Cursor IDE
Perplexity AI
Lovable
Builder.io
GitHub Copilot
day.js
Framer Motion

Recommendations for Future Teams

  • Refined Prompting: Use short, clean prompts to avoid bloated UI code, especially for complex state logic.
  • Manual Oversight: Take control over core, non-standard logic and validate all suggested dependencies.
  • Documentation: Document AI decisions, prompts that worked, and where AI broke down for future maintainability.

Background & Context

The VibeFlow Hackathon challenged developers to solve two critical pain points in modern education

Target Personas

Maya Chen

Student

A student struggling with specific concepts like machine learning, seeking a truly compatible and qualified mentor who understands her learning style.

Pain Point:

Finding compatible, qualified mentors for niche subjects

Dr. Aisha Patel

Senior ML Engineer

A senior ML engineer seeking an efficient way to mentor—needs comprehensive session management and student analytics without administrative overhead.

Pain Point:

Lack of efficient, dynamic platform for expert mentors

The EduVibe Platform Solution

EduVibe is an AI-powered EdTech system designed to connect students and mentors through personalized learning paths and robust session management.

Personalized Student Onboarding

Multi-part form capturing academic background, skill levels, learning style, and goals

AI-Driven Mentor Recommendation

Algorithm combining skill, availability, and goals with explainable AI (e.g., '90% match due to ML expertise')

Interactive Booking System

Context-aware time slot booking with real-time availability management

Dual Dashboard System

Student dashboard for sessions and progress; Mentor dashboard for analytics and availability

The Opportunity

Solving EdTech's personalization gap with AI—addressing the difficulty for students to find compatible, qualified mentors for niche subjects, and providing an efficient platform for expert mentors to share their knowledge.

Research Focus Areas

  • Where does AI truly accelerate development vs. create hidden friction?
  • How does AI impact code quality, security, and long-term maintainability?
  • What is the effectiveness of AI-driven state management in complex forms?
  • How do different AI tools perform across UI, backend, and research tasks?

Hackathon Timeline

Hackathon Launch

Teams begin building the EduVibe platform with AI tools

Development Phase

Participants use Claude, GPT-4o, V0 for UI, backend, and research

Submission Deadline

Teams submit their completed EduVibe implementations

Mentor Evaluation

Expert mentors review projects (50% Report, 30% Dev, 20% Presentation)

Report Analysis

Comprehensive analysis of AI tool effectiveness and workflows

Methodology

Our systematic approach to analyzing competitor tool usage and effectiveness

Evaluation Process

Gather Reports

Collect competitor submissions and tool usage data

Analyze Tools

Evaluate performance across different categories

Extract Insights

Identify patterns and key findings

Compare Results

Benchmark against VibeFlow capabilities

Market Research

  • Analyzed competitor submissions from the VibeFlow Competition
  • Evaluated tools based on usability, customization, performance, and cost
  • Collected quantitative metrics on completion rates and tool adoption

Competitor & Judge Feedback

  • Gathered insights from detailed case studies
  • Analyzed tool usage patterns, challenges, and outcomes
  • Judge panel emphasized customization and performance optimization

Data Collection

  • Extracted insights from AI tool reports (Claude, GPT, V0, etc.)
  • Analyzed architecture diagrams for system integration
  • Reviewed "WOW" and "FACEPALM" moments

Findings & Observations

Key insights into AI tool usage patterns and effectiveness across different development tasks

AI Accelerates Frontend/UI Development

AI tools significantly speed up UI scaffolding and basic component creation. Teams rapidly scaffolded entire multi-step onboarding flows with AI assistance.

Struggles with Animations & State

Complex animations and state management require significant human intervention. AI-generated carousel logic failed to track active slides correctly.

Asking Mode Outperforms Agent Mode for Logic

Interactive, step-by-step prompting (Asking Mode) was more effective for custom logic like Mentor Matching, while Agent Mode excelled at initial UI generation.

Claude & GPT-4o Excel in Different Domains

Claude 3 Opus valued for natural UI structuring; GPT-4o selected for high accuracy in generating explainable mentor matching output.

State Management Challenges

A primary focus was on front-end state management—ensuring real-time, consistent state updates and smooth transitions

State Loss

Occurred during form submissions, page refreshes, and state mismatches during navigation

Logical Errors

AI-generated carousel logic failed to track active slide correctly, requiring manual intervention

Over-Engineering

AI sometimes generated unnecessary form steps or bloated state logic that had to be refined

Tool Categories & Usage

Backend & Logic

Claude 3 Opus
GPT-4o
Claude 4 Sonnet

UI/Scaffolding

V0 by Vercel
Lovable
Builder.io

Research/Planning

Perplexity AI
Gemini Pro

IDE & Assistants

Cursor IDE
GitHub Copilot

Tools vs. Tasks Performance Matrix

Comparative analysis of AI tools across different development tasks

ToolUI ScaffoldingState ManagementBackend LogicMock DataResearch/Planning
Claude (Sonnet/Opus)
Medium
Low
High
High
High
GPT-4o/4.1
Medium
Medium
High
High
High
V0 by Vercel
High
Low
N/A
Medium
Low
Rocket.new
High
Low
N/A
Medium
Low
Builder.io
High
Medium
Low
Medium
Low
Lovable
High
High
Medium
High
Medium
Perplexity AI
Low
N/A
Low
N/A
High
GitHub Copilot
Medium
Medium
Medium
Medium
Medium
High
Excellent performance
Medium
Good performance
Low
Limited performance
N/A
Not applicable

WOW Moments

Where AI truly accelerated development

  • Rapid scaffolding of entire multi-step onboarding process
  • AI-generated explainable mentor recommendations (e.g., '90% match due to ML expertise')
  • Multi-part form field generation with basic validations
  • Simple CSS hover effects and transitions
  • Complex mock data generation for testing

FACEPALM Moments

Where AI created hidden friction

  • State loss during form submissions and page refreshes
  • Carousel active slide tracking failures
  • Over-engineered state logic requiring manual cleanup
  • Complex animated graphics replaced with simple placeholders
  • Outdated library suggestions (moment.js instead of day.js)

Key Insights

  • AI excels at structure and basic components but struggles with complex visual interpretation (animated graphics become simple placeholders)
  • Asking Mode (step-by-step prompting) is more effective for custom logic; Agent Mode is better for initial UI generation speed
  • Claude 3 Opus valued for natural UI structuring; GPT-4o selected for explainable AI output in mentor matching
  • V0 provides rapid UI scaffolding but with limited customization options
  • Dependency management is critical—AI may suggest outdated libraries (e.g., moment.js) that need manual replacement (e.g., day.js)

Analysis

Deep dive into tool performance patterns and competitive landscape insights

Competitor Tool Usage Analysis

Purpose & Use Case

Tools were primarily used for UI scaffolding, backend logic, and research tasks

Pros (Advantages)

Rapid prototyping, time savings, and automated code generation capabilities

Cons (Limitations)

Limited customization, state management issues, and manual intervention needs

Cost & Licensing

Varied pricing models from free tiers to enterprise subscriptions

Ease of Use

Generally accessible but with steep learning curves for advanced features

Impact & Outcomes

Mixed results with significant time savings offset by debugging overhead

Performance Analysis: Success vs. Failure Patterns

What Worked vs. What Failed

AreaWhat WorkedWhat FailedReason
Mock Data
GPT-4o and Claude made accurate test data with correct formatting
-
Clear data structure requirements
Validation Schemas
AI tools generated clean Zod/Yup schemas from clear input
-
Well-defined validation rules
Multi-step Forms
Logical form flows when given full form structure
State broke across steps
AI didn't track parent-child flow properly
Booking Logic-
Couldn't stop double bookings
Didn't understand database-level locks
Animations-
Buggy or clashed animations
AI missed timing or misused animation libraries
Backend Integration-
Frontend and backend didn't match
Field names and APIs didn't align correctly

What Worked Well

  • Mock data generation with accurate formatting
  • Validation schema creation from clear requirements
  • Multi-step form flows with proper structure
  • YAML to code translation for backend scaffolds
  • Planning from PRDs and requirement documents
  • Component reuse when provided with examples

What Failed

  • State management across complex form flows
  • Real-time booking logic and conflict resolution
  • Animation timing and library integration
  • Form field consistency and validation
  • Modular code architecture and reusability
  • Frontend-backend API alignment

Impact & Outcomes

Quantitative and qualitative assessment of AI tool effectiveness in the hackathon

AI Performance by Task Type (%)

Success vs failure rates across different development tasks

Key Insight: AI excels at UI scaffolding and mock data generation (85-90% success) but struggles significantly with state management and animations (25-35% success).

Development Outcome Distribution

Breakdown of AI-assisted development results

65%

Positive Outcomes

35%

Required Human Intervention

Positive Impacts

Time Savings

AI tools like V0 and Rocket.new reduced UI development time by 2-3 days, allowing teams to focus on core functionality and business logic.

Robust Algorithms

Fine-tuned mentor matching algorithms using Claude and OpenAI APIs significantly improved platform functionality and user experience.

Efficient Research

Perplexity streamlined requirements analysis and task planning, reducing research overhead by approximately 40%.

Negative Outcomes

Manual Interventions

Complex animations, state management, and backend logic required significant manual refactoring, often taking longer than building from scratch.

Bugs and Errors

Hydration errors in Next.js and incorrect project structures (e.g., Vite instead of Next.js) significantly increased debugging time.

Limited Customization

UI tools consistently failed to deliver brand-specific, responsive designs that matched project requirements and design systems.

Tools Used by Competitors

Comprehensive analysis of AI tools utilized during the hackathon, based on whitepaper findings

Tool Analysis Report

Detailed analysis of individual AI tools with strengths, weaknesses, and optimal use cases

AI Workflow: Agent vs Asking Mode

Agent Mode

Best for: Initial UI generation

Fast, autonomous generation of UI components and layouts

Asking Mode

Best for: Custom logic implementation

Interactive, step-by-step prompting for complex features like Mentor Matching Algorithm

Claude 3 Opus

Backend & Logic

Primary Task

Homepage (Task 1) & Mentor Dashboard (Task 5)

Use Case

Backend logic, multi-step flows, onboarding, validation, API/schema design

Strengths

  • Natural UI structuring
  • Verbose explanations
  • Clean TypeScript code
  • Step-by-step logic

Weaknesses

  • Slower response time for larger context
  • May overcomplicate simple logic
  • Needs strong prompts

Best For

Multi-step forms, backend validation flows, readable modular backend logic

Key Takeaway

Valued for natural UI structuring but slower for large context windows

GPT-4o

Backend & Logic

Primary Task

Mentor Matching Algorithm (Task 3)

Use Case

Full-stack scaffolding, mock data, validation schemas, explainable AI output

Strengths

  • High accuracy in explainable output
  • Fast processing
  • Handles long inputs
  • Clean schema generation

Weaknesses

  • Occasionally mismatched domain expertise with availability
  • Can break multi-step logic
  • False confidence in outputs

Best For

Mentor matching with explainability (e.g., '90% match due to ML expertise')

Key Takeaway

Selected for high accuracy in generating explainable AI recommendations

V0 by Vercel

UI Generation

Primary Task

UI Generation & Dashboard Preview

Use Case

Text or Figma to responsive apps using React, Next.js, Tailwind, shadcn/UI

Strengths

  • Rapid UI scaffolding
  • Fast initial generation
  • Design-to-code
  • One-click Vercel deploy

Weaknesses

  • Limited customization options
  • Needs manual tweaks for complex logic
  • Struggles with complex animations

Best For

Rapid UI prototypes, dashboard previews, app layout drafts with real framework code

Key Takeaway

Used for rapid UI scaffolding but limited customization

Cursor IDE

IDE/Agents

Primary Task

AI-Assisted Development

Use Case

AI-powered code editor for writing, refactoring, managing code with multiple model support

Strengths

  • Deep project understanding
  • Agent Mode for initial UI
  • Asking Mode for custom logic
  • Supports GPT, Claude, Gemini

Weaknesses

  • Needs good prompts for accurate output
  • Can misinterpret vague prompts
  • Learning curve for .cursorrules

Best For

Full-stack development, debugging, AI-assisted code generation with model flexibility

Key Takeaway

Asking Mode preferred for custom logic; Agent Mode for rapid UI generation

Perplexity AI

Research & Planning

Primary Task

Requirements Analysis & Task Planning

Use Case

Planning tool, breaks down requirements, task sequencing, not a code generator

Strengths

  • Strong research and summarization
  • Great for structuring tasks
  • Endpoint planning
  • Real-time web search

Weaknesses

  • Cannot write backend logic
  • Outputs are high-level only
  • No code generation capability

Best For

Planning backend architecture, mapping tasks, researching system designs before coding

Key Takeaway

Excellent for pre-development planning and requirement analysis

Lovable / Builder.io

UI Generation

Primary Task

Full-Stack App Generation

Use Case

Generates full apps from text or Figma with backend integrations

Strengths

  • Clean code output
  • Rich integrations (Supabase, Stripe)
  • Mobile + web support
  • Full-stack generation

Weaknesses

  • Complex UI logic needs editing
  • May produce verbose code
  • Limited design control

Best For

MVPs, internal tools, dashboards when time is short

Key Takeaway

Good for rapid full-stack prototypes but requires refinement

SWOT Analysis

Strategic analysis of strengths, weaknesses, opportunities, and threats for each AI tool

SWOT Analysis Matrix

Tool
Strengths
Weaknesses
Opportunities
Threats
Claude (Sonnet / Opus)Strong logic, clean code, great for validation and auditsSlower, may over-engineer, needs clear promptsUse in audits, backend flows, onboardingHigh cost, slower vs GPT-4o
GPT-4o / GPT-4.1Fast, large context, great for mock data and APIsHallucinates logic, broken states, false confidenceScaffolding, validation chains, dashboardsMisleading output, loss of trust
V0Text-to-code UI, Figma integrationManual fixes for animations, complex UISpeed up MVP UIs, collaborative workflowsAI outputs need heavy edits
Rocket.newNatural language to full-stack app, high-quality codeVerbose code, needs customizationNon-devs ship MVPs faster, time-savingHard to debug complex flows
Perplexity AIExcellent planning, summarization, task breakdownNo code generation, high-level only, no memoryPlanning before AI code genReplaced by tools that also write code
Cursor IDEInline AI coding, .cursorrules, multi-model, Git integrationMisinterprets vague prompts, config learning curveRaise code quality with inline rulesErrors if used blindly, refactor time adds up

Recommendations & Improvements

Strategic guidelines for optimizing AI tool usage and development workflows

Core Recommendation

AI is indispensable for rapid prototyping and generating standard components. However, its use requires constant human validation, particularly for state management and complex decision logic.

Prompting Best Practices

Tool Selection Guidelines

Mentor Matching Logic

Recommended
GPT-4o for explainable AI recommendations

High accuracy in generating explainable output (e.g., '90% match due to ML expertise')

Alternatives:

Claude 3 Opus
Claude 4 Sonnet

UI Structure & Dashboards

Recommended
Claude 3 Opus for natural UI structuring

Valued for verbose explanations and clean component organization

Alternatives:

V0 by Vercel
GPT-4o

Rapid UI Prototyping

Recommended
V0 by Vercel for fast scaffolding

Extremely fast initial generation, though limited customization

Alternatives:

Lovable
Builder.io

Custom Logic Implementation

Recommended
Cursor IDE with Asking Mode

Interactive prompting prevents hardcoding issues in complex features

Alternatives:

Direct Claude/GPT API

Critical Human Review Areas

Developers must take manual control over these non-standard logic areas

State Management

High

Validate state persistence across form submissions, page refreshes, and navigation

Animation Logic

High

Review AI-generated animations—carousel tracking and transitions often fail

Dependency Validation

Medium

Check all AI-suggested libraries for deprecation (e.g., moment.js → day.js)

Complex Visual Designs

Medium

AI struggles with non-standard geometric patterns; manual implementation required

Documentation Requirements

Detailed documentation is crucial for engineering quality and future maintainability

1
Which AI tools were used for each feature
2
Which prompts worked best and why
3
Where the AI broke down and manual intervention was needed
4
Library replacements made (e.g., moment.js → day.js)
5
State management decisions and edge cases handled

VibeFlow Platform Advantages

The proposed solution addresses key pain points discovered during the hackathon

Advanced UI generation with seamless animations
Centralized state management interface
Robust backend integration (booking systems, conflict resolution)
Optimized prompting for reduced manual intervention
Built-in code integrity checks and validation
Definitive benchmark for AI in software development