Bloom.AI

This is the FIRST VERSION. The team has taken this project to scale: https://linktr.ee/bloom.ai

Track: Heart of the Matter, Challenges: ROBLOX, Google Gemini

Inspiration

What if your words and actions had the power to shape the digital world around you?

Not metaphorically, but visibly and instantly-where kindness blooms and negativity withers?

What if this power could also cultivate personal wellbeing and stronger communities?

As our lives move increasingly online, the need for civility, empathy, and genuine connection is critical. Yet, fostering these qualities, especially among young audiences, is challenging. Online negativity, bullying, and the general stress of digital interaction can overshadow the potential for positive connection, leaving many feeling disconnected or unsafe. We saw an opportunity to address this directly.

That's why we built Bloom.AI, a Roblox experience designed to decrease stress, combat online negativity, and actively cultivate civility. It's a world where players nurture their virtual environments and their own wellbeing through positive interactions, collaborative action, and the simple act of planting.

We uniquely harness AI and engaging gameplay to turn kindness into tangible growth – restoring environments, unlocking new possibilities, and fostering crucial social-emotional skills. It’s stress relief, anti-bullying, and collaborative world-building combined into one ever-evolving adventure.

The Problem

In most online spaces, the effects of our words are invisible or abstract. This invisibility enables toxic behaviors-bullying, exclusion, negativity-to spread, contributing to online stress and harming community health. Even well-meaning users struggle when kindness feels unrewarded and the consequences of negativity aren't clear. The digital world often lacks inherent mechanisms that visualize impact and incentivize positive, collaborative behavior.

These key challenges sparked a critical question:

How might we make the impact of digital communication-both positive and negative-immediately visible and meaningful, fostering civility, reducing stress, and teaching social-emotional skills?

Which became a guiding inquiry that led us to our core hypothesis:

How can we design an engaging virtual environment where every word and collaborative action directly transforms the world, providing instant feedback that reinforces positive behavior and discourages negativity?

Prototyping Process

Objective Our primary goal during prototyping was to rapidly validate our core hypothesis. We needed to test if seeing words make plants bloom, negativity cause withering, and collaboration restore structures provided meaningful feedback loops.

Approach We adopted a rapid prototyping strategy focused on bridging the gap between ideation and a testable experience, concentrating on the central mechanics:

Processing chat sentiment (using AI) to influence the environment and grant rewards.
Enabling collaborative environmental actions (cleaning, basic building) for points/rewards.
Using earned points/resources to plant, grow, and restore.
Visualizing the environmental impact (growth, decay, restoration) in real-time.
Simulating basic perspective-taking or intervention scenarios.

Phase 1: Visual Concept & Asset Ideation To explore the visual language of Bloom AI-how flourishing life and decaying neglect would manifest-we utilized Roblox Cube. As Roblox's foundational generative AI system for 3D, Cube allowed us to:

Rapidly generate diverse ideas for 3D assets (plants, environmental elements, structures) that could represent positive growth or negative impact.
Visualize how scenes might transform based on simulated positivity scores. This helped us establish potential visual styles and asset types for the dynamic environment before building complex systems, aligning with the vision of AI supporting developer creativity in generating Roblox experiences.

Phase 2: Interactive Mechanic Prototyping Translating text input into immediate visual environmental feedback required a functional prototype. We acknowledged that a major bottleneck in development is often the delay between concept and interaction testing. Therefore, to accelerate this crucial phase, we chose to utilize Lovable.dev. Its key advantages directly addressed our needs:

Rapid Transformation from Idea-to-Prototype: Lovable.dev enabled us to instantly translate our core concept (words → world change) into an interactive prototype.
Focus on Core Loop: This allowed us to quickly build, test, and iterate on the user experience of the sentiment-visual feedback system without getting bogged down in the complexities of full Roblox engine development or manual HTML/CSS/JS coding for this specific stage.

How We Built It

Bloom.AI integrates a Roblox game experience with real-time AI sentiment analysis and a supporting dashboard, creating a system where positive actions visibly shape the world, foster wellbeing, and teach vital skills.

1. Roblox Game: Bloom AI

Bloom.AI’s Roblox game was architected using our PriorityEngine, a simple yet powerful modular framework designed to streamline and organize the initialization and management of game components. This approach sets Bloom.AI apart from typical Roblox games by prioritizing maintainability, scalability, and reliability from the ground up via:

Module Initialization: Foundational modules (such as utility libraries, data structures, and shared services) are loaded first. These modules are reusable and do not directly interact with the game’s client or server scripts but provide essential services required by controllers.
Controller Loading: After modules are ready, controllers-scripts that handle the game’s logic and interactions-are loaded. Controllers are divided into client-side (UI, input, local effects) and server-side (game state, player data, backend logic) categories, respecting Roblox’s networking model

the game experience is also built around interconnected systems:

Words Shape Worlds (Sentiment & Environment): Player chat is processed in real-time by Gemini LLM to assess positivity.
Visual Feedback: Positive communication generates a subtle "Positivity Aura," causing small plants to bloom nearby and contributing to the overall health of the area. Consistent positivity helps larger plants flourish and structures begin to restore. Conversely, negativity can cause visible withering and decay.
Resource Generation: Positive chat generates points or resources needed for planting and restoration.
Collaborative Action for Restoration: Players must work together to improve their environment:
- Cleaning: Cooperatively cleaning litter and pollution provides points and visibly improves the immediate area.
- Building/Restoration: Larger restoration projects (fixing fountains, rebuilding structures) require teamwork, potentially enhanced by positive communication during the task, yielding greater rewards and unlocking new visual elements or minor area functionalities.
Stress-Reducing Planting & Growth: Players use earned points/resources to plant trees and cultivate gardens. This core activity provides a calming experience, directly linking positive behavior and collaboration to tangible, vibrant outcomes.
Progression & Unlocking: Growing plants and restoring the environment generate "Life Force Points." These points might determine access to new areas or dimensions or unlock the ability to purchase different types of seeds, tools, or cosmetic items.

Social-Emotional Learning Mechanics:

Bullying Intervention Scenarios: Scripted events present players with opportunities to practice different strategies (support, direct address, reporting) for dealing with simulated bullying, with feedback on effectiveness.
Perspective Shift Challenges: Special challenges allow players to temporarily experience the game from another's viewpoint, building empathy by highlighting different perceptions or feelings in a given situation.

PriorityEngine ensures these systems load reliably, managing dependencies so that AI analysis, environmental state, and player progression systems function cohesively.

2. Sentiment Analysis Dashboard

This web-based tool (Next.js, TailwindCSS, shadcn/ui) complements the game:

Live Chat Monitoring & Sentiment Scoring: Streams player chat with real-time Gemini LLM sentiment scores, making communication impact visible for moderation and analysis.
Sentiment Analyzer: Allows manual testing of messages via Gemini.
Community Analytics: Aggregates data on message volume, sentiment trends, and player participation, offering insights into community health.
Leaderboard: Ranks players by cumulative positivity scores/contributions, encouraging sustained positive engagement.

Challenges We Ran Into

Developing Bloom AI involved navigating significant technical and design hurdles:

Real-time Data Integration: Creating a robust, low-latency pipeline between Roblox (chat, actions) and our backend (Flask, Gemini API) for real-time environmental updates and scoring was complex.
Nuanced Sentiment Scoring & Balancing: Defining a meaningful "positivity score" framework (-100 to +100) using Gemini that accurately reflected diverse communication styles and translated into fair, motivating game mechanics (rewards, environmental impact) required extensive tuning.
Performant AI & Environmental Systems: Integrating Gemini LLM for real-time analysis and managing the dynamic environmental transformations (growth, decay, restoration effects) without impacting Roblox client performance was critical.
Designing Effective SEL Mechanics: Creating engaging and genuinely educational scenarios for bullying intervention and perspective-taking within the game context required careful design and iteration.

Accomplishments We're Proud Of

Despite the challenges, we achieved significant milestones:

Building a Full, Immersive Game: We created a complete, engaging Roblox experience realizing our core vision: a world where players visibly see their words and actions matter, fostering wellbeing through planting and collaboration, while learning civility through integrated mechanics.
Integrating Real-time AI for Gameplay & Environment: The seamless integration of real-time sentiment analysis (Gemini LLM) directly influencing both player rewards and the dynamic state of the virtual world is a key technical achievement.
Implementing Foundational SEL Mechanics: Successfully incorporating initial versions of mechanics designed to teach empathy and positive intervention strategies within the gameplay loop.
Developing a Comprehensive Dashboard: Providing valuable real-time insights and moderation tools that support the game's goals of fostering a positive community.

What's Next for Bloom.AI

Bloom.AI is a foundation for continued growth. We plan for these future enhancements:

Deeper Environmental Interaction & Progression: Expanding the world restoration mechanics, potentially unlocking new visually distinct areas or dimensions based on collective community positivity and collaborative achievements, fulfilling the original "unlocking" concept.
Advanced Content Moderation & Civility Recognition: Integrating more sophisticated AI (refining Gemini usage, potentially other models) to detect nuanced harmful content and proactively identify/reward exemplary civil communication and collaboration.
Richer SEL Scenarios: Expanding the library and complexity of bullying intervention and perspective-taking scenarios, offering more diverse learning opportunities.
AI Agent Integration: Incorporating AI agents to support the community:
- An AI Moderator to assist human moderators in identifying issues.
- An AI Mediator to offer constructive suggestions during player conflicts.
- A Policy Agent (using RAG) to answer questions about Roblox's rules and Bloom AI's community standards.
- AI agents providing Admin suggestions and relevant educational resources to players based on observed interactions.