diff --git a/src/app/ai.service.ts b/src/app/ai.service.ts
new file mode 100644
index 0000000..7f4c0d2
--- /dev/null
+++ b/src/app/ai.service.ts
@@ -0,0 +1,144 @@
+import * as tf from "@tensorflow/tfjs";
+
+export class AiService {
+
+    constructor() {
+
+    }
+
+private currentModel;
+flipX(arr) {
+  return [arr.slice(6), arr.slice(3, 6), arr.slice(0, 3)].flat();
+};
+
+flipY(arr) {
+
+
+ this.flipX(arr.slice().reverse());
+}
+// Creates a 1 hot of the diff
+showMove(first, second) {
+  let result = [];
+  first.forEach((move, i) => {
+    result.push(Math.abs(move - second[i]));
+  });
+  return result;
+};
+
+getMoves (block) {
+  let x = [];
+  let y = [];
+  // Make all the moves
+  for (let i = 0; i < block.length - 1; i++) {
+    const theMove = this.showMove(block[i], block[i + 1]);
+    // Normal move
+    x.push(block[i]);
+    y.push(theMove);
+    // Flipped X move
+    x.push(this.flipX(block[i]));
+    y.push(this.flipX(theMove));
+    // Inverted Move
+    x.push(block[i].slice().reverse());
+    y.push(theMove.slice().reverse());
+    // Flipped Y move
+    x.push(this.flipY(block[i]));
+    y.push(this.flipY(theMove));
+  }
+  return { x, y };
+};
+
+constructModel() {
+  this.currentModel && this.currentModel.dispose();
+  tf.disposeVariables();
+
+  const model = tf.sequential();
+
+  model.add(
+    tf.layers.dense({
+      inputShape: [9],
+      units: 64,
+      activation: "relu"
+    })
+  );
+
+  model.add(
+    tf.layers.dense({
+      units: 64,
+      activation: "relu"
+    })
+  );
+
+  model.add(
+    tf.layers.dense({
+      units: 9,
+      activation: "softmax"
+    })
+  );
+
+  const learningRate = 0.005;
+  model.compile({
+    optimizer: tf.train.adam(learningRate),
+    loss: "categoricalCrossentropy",
+    metrics: ["accuracy"]
+  });
+
+  this.currentModel = model;
+  return model;
+};
+
+ getModel() {
+  if (this.currentModel) {
+    return this.currentModel;
+  } else {
+    return this.constructModel();
+  }
+};
+
+async trainOnGames(games, setState) {
+  const model = this.constructModel();
+  // model.dispose();
+  let AllX = [];
+  let AllY = [];
+
+  // console.log("Games in", JSON.stringify(games));
+  games.forEach((game) => {
+    AllX = AllX.concat(game.x);
+    AllY = AllY.concat(game.y);
+  });
+
+  // Tensorfy!
+  const stackedX = tf.stack(AllX);
+  const stackedY = tf.stack(AllY);
+  await this.trainModel(model, stackedX, stackedY);
+
+  // clean up!
+  stackedX.dispose();
+  stackedY.dispose();
+
+  setState(model);
+  // return updatedModel;
+};
+
+async trainModel (model, stackedX, stackedY) {
+  const allCallbacks = {
+    // onTrainBegin: log => console.log(log),
+    // onTrainEnd: log => console.log(log),
+    // onEpochBegin: (epoch, log) => console.log(epoch, log),
+    onEpochEnd: (epoch, log) => console.log(epoch, log)
+    // onBatchBegin: (batch, log) => console.log(batch, log),
+    // onBatchEnd: (batch, log) => console.log(batch, log)
+  };
+
+  await model.fit(stackedX, stackedY, {
+    epochs: 100,
+    shuffle: true,
+    batchSize: 32,
+    callbacks: allCallbacks
+  });
+
+  console.log("Model Trained");
+
+  return model;
+};
+
+}
\ No newline at end of file