Two batches of action-arcade games spanning Journeys 2–8. Each makes the kid an AI engineer. Click any tile to play. 🔊 toggle in the top-right mutes game sounds.
A neural network fires neurons in sequence — watch the ZAP! then tap them back in the same order. Teaches how neural nets activate.
A sequence shows on Codi's prediction crystal with one piece blank. Tap the right fill before the orb dims. Teaches the foundational AI skill: spotting patterns.
A dataset rolls in — one row is mislabeled. Tap the bad row before it ships. Teaches data quality & anomaly detection in green-terminal style.
Codi's AI brain shows 3 training examples (🐶→Animal, 🍎→Food, 🚗→Vehicle). Then new items arrive — classify them the way the AI would. Teaches supervised learning.
A fuzzy image comes into focus inside Codi's vision terminal. Tap the right label before it's fully clear. Teaches image recognition.
Build the AI's prompt by picking a STYLE, SUBJECT and SETTING word, then hit Generate. Match the target painting. Teaches prompt engineering.
Codi dashes across a sunset-purple city skyline collecting AMAZING/LOVE word coins and dodging boring/UGH blocks. Teaches sentiment analysis.
Word bubbles drift up through a sherbet sky. Tap only the happy ones — tapping a grumpy bubble costs a heart. Teaches sentiment classification.
In Codi's AI courtroom, decide if each AI decision is FAIR or BIASED before the gavel falls. Relatable kid-friendly scenarios. Teaches bias awareness.
Codi built a robot. You write its program. Stack ↑↓←→ commands, hit RUN, navigate past walls to the flag. Teaches algorithmic thinking and sequencing.
An AI product needs PM, ML engineer, UX designer and Domain expert. Send each task to the right teammate's column. Teaches AI team roles.
A Codi-pet does something each round. Tap REWARD 👍 or NOPE 👎 based on whether the behavior is good. Teaches reinforcement learning by feedback.
A user texts the chatbot. Pick the best reply from 3 options in a phone-chat UI. Teaches conversational AI & context.
4 photo cards on the evidence board. One is the AI-generated fake — look for the chimera tell. Teaches spotting AI-generated images.
Top-down road. Obstacles appear (🚸 kid crossing, 🚦 light, 🦌 deer). Pick STOP / SLOW / GO before the car gets there. Teaches AI in self-driving cars.
A sentence with 3 blanks. Pick noun + verb + place words and hit WEAVE. The AI generates the story — match the target. Teaches text-generation prompts.
The AI lab shows 4 labels with confidence bars (cat 78%, dog 12%…). Tap the AI's top guess. Teaches AI uncertainty & probability.
A user profile shows 3 likes. Pick the best show recommendation from 3 streaming-app cards. Teaches Netflix/YouTube-style recommendation algorithms.
Classify each item (Animal / Plant / Object). Correct answers grow flowers in your garden strip. Wrong ones wilt them. Teaches iterative training.
Codi's first day at the AI Academy. The mascot is missing. The school's AI is broken. Codi, Pixel and Professor Lin work through a real pattern-recognition puzzle — and learn that a pattern is a question, not an answer.
A neural network fires neurons of different colors. The TARGET shows which one to catch — slide the catcher to grab the right color before it lands. 5 in a row triggers Mega AIGo mode.
A low-poly object spins on a hologram pedestal. The AI shows 4 candidate labels with confidence bars — but confidence bars can lie. Pick the label that matches what you actually see.
Drops fall from a faucet — each one a data point that needs to be sorted. Click paddles to flip their tilt and bounce each drop into the matching bin (Animal / Vehicle / Food).
Two mandalas sit framed like paintings. Tune four parameters (Hue / Density / Shape / Roughness) to make your generation match the target. Teaches that AI creativity is parameter tuning, not magic.
Codi the robot wants to reach a magnet but hates spikes. Place tasty cookies on safe tiles to bias his path. Click Train Codi — he uses BFS pathfinding biased by your rewards.
A 3D galaxy of words clustered by meaning. The AI gives you a new word — pick which cluster it belongs to. Teaches embeddings: similar concepts sit close in semantic space.
A real balance beam with hanging pans. Drop data samples on the lighter side to keep training data balanced. Tilt past 8° = biased AI = audit fails.
Four procedurally generated faces. One has a subtle AI 'tell' — asymmetric eyes, off-center mouth, hair float, hue mismatch. Spot it before the timer runs out.
Codi's autonomous car drives forward. Hazards approach — kid, cyclist, leaf. Pick STOP / SLOW / GO before each crosses the decision line. Teaches AV sensor fusion.
You sit at the center of a 3D taste-cloud. Three candidate items at varying distances. Pick the closest — that's how AI recommenders work (collaborative filtering).
A real Plinko peg board. Bet on a bin (1–5), drop the ball, watch stochastic physics decide. Center is statistically most likely, but anything can happen. Teaches AI uncertainty.
Retro CRT oscilloscope. A signal scrolls across with anomalies (spikes, dips, flatlines). Click them before they scroll off. AI's dashed confidence band shows the normal range.
Phone messaging UI. A customer sends a message — tap the matching intent (greeting / question / complaint / thanks). Before a chatbot can reply, it has to know what the user wants.
Isometric 3D warehouse. Click a goal tile — watch the AI's BFS frontier expand tile-by-tile, then the robot follows the shortest path. The first true visualization of how AI 'thinks'.
Scientific chalkboard with real loss curves (W-shape, hidden trough…). Pick a starting θ, drop the ball, watch gradient descent settle. Find the global minimum, beware local ones.
Neon 8-step × 5-pitch sequencer. AI plays a melody — replicate it by tapping cells. Real Web Audio notes (C4 to C5). Teaches that AI music is sequence pattern matching.
Pirate treasure cove with 5 hidden-reward chests. Should you keep clicking your favorite, or try a new one? Explore vs exploit — the heart of every RL algorithm.
6 client devices around a central AI. Each sends a transmission — blue=safe encrypted gradient, red=raw data leak. Approve safe ones, block leaks. Privacy must hold.
Brass grandfather-clock pendulum. Watch it swing for 2s, then a curtain falls. Pick where the tip will be at the reveal. Double pendulums = chaos = limits of AI forecasting.
A pixel-art cat. The AI sees CAT (high confidence). Paint a few colored noise pixels — watch the AI flip its prediction to BANANA. Humans still see a cat. AI is fragile to tiny noise.