SkillRacer

Inspiration

The modern learning ecosystem is losing a war against the dopamine economy. We realized that traditional "EdTech" and passive study apps simply cannot compete with the variable-reward loops of TikTok or AAA video games. We wanted to cure the "doom-scrolling" epidemic, but we knew we couldn't do it by building another boring flashcard app. We needed to weaponize the exact same behavioral psychology and high-octane mechanics used by the gaming industry, but redirect that addiction toward high-value software engineering skills.

Furthermore, we recognized a massive gap in how platforms handle "Cognitive Fatigue." Students burn out in isolation. We wanted to build a system that doesn't just grade you, but actively monitors your mental redline using implicit gameplay telemetry.

What it does

SkillRacer is a cyberpunk-themed, real-time multiplayer educational esport. It transforms coding mastery into a high-stakes racing and raiding ecosystem:

RPM Racer (PvP): A high-speed drag race where your car's acceleration is directly tied to your coding reflexes. "Perfect Shifts" (answering fast) deploy nitrous; wrong answers stall your engine.
Alien Raid (4-Player PvE): A cooperative boss battle. Four pilots defend a server from an AI UFO. You shoot the boss by answering correctly. If you answer wrong, the boss targets you.
The Command Center: A data-rich esports profile featuring a GitHub-style Activity Matrix, social "Blood Oath" peer streaks, and a digital Vault of minted skills.
Pilot Vitals (The Cognitive Orthotic): A background system that tracks millisecond hesitation and error rates to predict burnout, intervening with an AI Copilot when you are "redlining."

How we built it

We built a highly scalable, real-time architecture:

Frontend: Next.js (App Router) with Tailwind CSS for a dark, aggressive, photo-realistic Cyberpunk UI, utilizing complex Framer Motion and CSS animations for the racing visuals.
Backend: A Node.js server acting as the game engine, utilizing Socket.io for ultra-low latency WebSocket synchronization of player coordinates, Boss HP, and multiplayer lobbies.
Database: MongoDB Atlas to handle complex user relations and massive streams of high-frequency gameplay telemetry.
AI Brain: Google's Gemini 1.5 Pro, integrated directly into the Node.js backend to serve as the live Dungeon Master, the Code Judge, and the Behavioral Analyst.

NRVE Track

SkillRacer is tailor-made for the NRVE challenge. We completely eliminate isolated doom-scrolling by replacing it with Peer-to-Peer Group Learning. Our "Alien Raid" mode forces 4 strangers to collaborate to defeat a coding boss. But the real magic is our Feynman Chat Arbitration. When a player "dies" in the raid, they enter Spectator Mode. To revive, they must use plain English (the Feynman Technique) in the chat to explain the concept they just failed to their surviving teammates. Our AI instantly evaluates their explanation, and if it helps the team, they are revived. Furthermore, our "Blood Oath" feature links users' daily learning streaks together, weaponizing social accountability to ensure continuous engagement.

Dallas AI Track

We addressed the Dallas AI prompt by creating a hyper-personalized, continuously adaptive learning journey. Traditional platforms offer static paths. SkillRacer uses a Flow State Calibrator. Our AI continuously monitors a user's reaction time; if they are answering too quickly, the system dynamically mutates the code complexity mid-race to keep them in the optimal zone of proximal development. Additionally, our "Career Oracle" dashboard analyzes a user's yearly learning radar charts and provides daily, personalized agentic insights on what specific micro-skills they need to target to remain competitive.

Use of Gemini

Gemini is not a chatbot in SkillRacer; it is the core gameplay engine. We utilized Gemini to its maximum potential across four distinct vectors:

The AI Dungeon Master: In Alien Raids, Gemini queries the skill profiles of the 4 matched players, finds their shared weaknesses, and dynamically generates the specific Python bugs the UFO will attack them with.
The Feynman Judge: We use Gemini's natural language processing to act as a real-time semantic gatekeeper, evaluating whether a dead player's plain-English explanation of syntax demonstrates true conceptual mastery to trigger a game-state revival.
Multimodal "Pit Stops": When a user fails repeatedly, Gemini instantly processes YouTube video tutorials, extracts the exact 30-second visual snippet relevant to their failure, and plays it on screen.
Massive Context RAG Reporting: At the end of the week, we feed thousands of rows of raw MongoDB keystroke telemetry and chat transcripts into Gemini's massive context window. Gemini acts as a behavioral analyst, outputting a clinically-styled PDF diagnostic predicting the user's cognitive burnout risk.

Use of MongoDB

MongoDB is the anchor that makes our real-time tracking possible. We pushed Atlas to its limits:

Time-Series Collections: We ingest high-frequency, millisecond-precision telemetry from the drag races (hesitation times, keystroke speeds, error rates). This requires the massive write-speed capabilities of MongoDB Time-Series to accurately track the biometric data needed for our burnout predictions.
Atlas Vector Search: We convert user knowledge gaps into mathematical vector embeddings. When creating a 4-player Alien Raid lobby, we use Vector Search to instantly match players who have perfectly complementary skill deficiencies, ensuring balanced teams.
Complex Document Structures: We used rich JSON documents to store the interconnected "Blood Oath" peer streaks and the massive nested arrays of our 30-level dynamic game tracks.

Use of Google Antigravity

To hit the ground running and maximize our engineering time, we leveraged Google Antigravity as our primary rapid-prototyping engine. Building a data-heavy esports dashboard and a real-time multiplayer interface from scratch would normally consume the entirety of a hackathon weekend. Antigravity allowed us to completely bypass the grueling boilerplate phase. We used highly detailed architectural prompts to generate our Next.js starter templates, complex responsive layouts, and the foundational data visualization components for our "Pilot Vitals" tracking. By offloading the initial component scaffolding to Antigravity, we bought ourselves the crucial hours needed to manually engineer the hard stuff: fine-tuning the React lifecycle states, injecting our aggressive cyberpunk aesthetic, and building the complex Node.js WebSocket and Gemini RAG integrations that power the actual game loop.

Challenges we ran into

Synchronizing real-time Socket.io game states with asynchronous LLM API calls was incredibly difficult. If Gemini took 2 seconds to evaluate a "Feynman" chat explanation, the game loop couldn't freeze. We had to build complex, non-blocking asynchronous queues in Node.js to ensure the UFO kept firing while the Oracle evaluated the dead player's response. Additionally, writing prompt constraints strict enough to force Gemini to return perfectly structured JSON for our game engine to parse—without hallucinating extra conversational text—required dozens of iterations.

Accomplishments that we're proud of

We successfully transformed "learning Python syntax" into an adrenaline-pumping experience. The moment we got 4 players in a WebSocket lobby, and the Gemini-powered UFO started dynamically targeting the player who missed a List Comprehension question, we knew we had built something special. We proved that you don't need hardware to track cognitive load; you can do it purely through software telemetry and agentic AI.

What we learned

We learned the profound difference between wrapping a UI around an LLM and actually integrating an LLM into an application's core logic loop. We mastered Node.js WebSocket room management, the intricacies of MongoDB Time-Series data, and the power of "Behavioral RAG"—using AI to find patterns in raw database logs rather than just querying text documents.

What's next for SkillRacer

We plan to expand the "Vault" to include System Design and Cloud Architecture (AZ-900) tracks. We want to introduce WebXR to bring the Alien Raid into immersive VR, and eventually host live, sponsored SkillRacer tournaments where university students compete in high-speed, live-streamed coding drag races.