diff --git a/.jules/bolt.md b/.jules/bolt.md
new file mode 100644
index 0000000..e69de29
diff --git a/packages/ai/src/index.test.ts b/packages/ai/src/index.test.ts
new file mode 100644
index 0000000..08e2af7
--- /dev/null
+++ b/packages/ai/src/index.test.ts
@@ -0,0 +1,61 @@
+
+import { describe, it, expect } from "vitest";
+import {
+  generateEmbeddings,
+  cosineSimilarity,
+  normalizeEmbeddingText,
+  embeddingContentHash,
+  embeddingDimensions
+} from "./index.js";
+
+describe("AI Embedding Utils", () => {
+  it("should normalize text correctly", () => {
+    expect(normalizeEmbeddingText("  hello   world  ")).toBe("hello world");
+  });
+
+  it("should generate consistent hashes", () => {
+    const text = "test message";
+    const hash1 = embeddingContentHash(text);
+    const hash2 = embeddingContentHash(text);
+    expect(hash1).toBe(hash2);
+    expect(hash1).toMatch(/^[a-f0-9]{64}$/);
+  });
+
+  it("should generate deterministic synthetic embeddings", async () => {
+    const text = "deterministic test";
+    const result1 = await generateEmbeddings([text], { forceSynthetic: true });
+    const result2 = await generateEmbeddings([text], { forceSynthetic: true });
+
+    expect(result1[0].values).toEqual(result2[0].values);
+    expect(result1[0].dimensions).toBe(embeddingDimensions);
+  });
+
+  it("should produce normalized synthetic vectors", async () => {
+    const text = "normalized test";
+    const [record] = await generateEmbeddings([text], { forceSynthetic: true });
+
+    const magnitude = Math.sqrt(
+      record.values.reduce((sum, val) => sum + val * val, 0)
+    );
+    // Should be very close to 1
+    expect(magnitude).toBeGreaterThan(0.999999);
+    expect(magnitude).toBeLessThan(1.000001);
+  });
+
+  it("should calculate cosine similarity correctly", () => {
+    const v1 = [1, 0, 0];
+    const v2 = [1, 0, 0];
+    const v3 = [0, 1, 0];
+    const v4 = [-1, 0, 0];
+
+    expect(cosineSimilarity(v1, v2)).toBeCloseTo(1);
+    expect(cosineSimilarity(v1, v3)).toBeCloseTo(0);
+    expect(cosineSimilarity(v1, v4)).toBeCloseTo(-1);
+  });
+
+  it("should handle empty or mismatched vectors in cosineSimilarity", () => {
+    expect(cosineSimilarity([], [1, 2, 3])).toBe(0);
+    expect(cosineSimilarity([1, 2], [1, 2, 3])).toBe(0);
+    expect(cosineSimilarity(undefined, [1])).toBe(0);
+  });
+});
diff --git a/packages/ai/src/index.ts b/packages/ai/src/index.ts
index 6eb88fc..111b465 100644
--- a/packages/ai/src/index.ts
+++ b/packages/ai/src/index.ts
@@ -15,21 +15,33 @@ const normalizeText = (value: string) => value.replace(/\s+/g, " ").trim();
 const contentHashFor = (value: string) =>
   crypto.createHash("sha256").update(normalizeText(value)).digest("hex");
 
-const seededUnitValue = (seed: string, index: number) => {
-  const digest = crypto.createHash("sha256").update(`${seed}:${index}`).digest();
-  const int = digest.readUInt32BE(0);
-  return int / 0xffffffff;
+/**
+ * Fast, deterministic PRNG based on Mulberry32.
+ * @param seed - 32-bit integer seed
+ */
+const mulberry32 = (seed: number) => {
+  return () => {
+    /* biome-ignore lint/style/noParameterAssign: optimization */
+    seed |= 0;
+    /* biome-ignore lint/style/noParameterAssign: optimization */
+    seed = (seed + 0x6d2b79f5) | 0;
+    let t = Math.imul(seed ^ (seed >>> 15), 1 | seed);
+    t = (t + Math.imul(t ^ (t >>> 7), 61 | t)) ^ t;
+    return ((t ^ (t >>> 14)) >>> 0) / 4294967296;
+  };
 };
 
 const syntheticVector = (text: string, dimensions = DEFAULT_EMBEDDING_DIMENSIONS) => {
   const normalized = normalizeText(text);
-  const hash = contentHashFor(normalized);
-  const values = Array.from({
-    length: dimensions
-  }, (_, index) => {
-    const centered = seededUnitValue(hash, index) * 2 - 1;
-    return Number(centered.toFixed(8));
-  });
+  const hash = crypto.createHash("sha256").update(normalized).digest();
+  const view = new DataView(hash.buffer, hash.byteOffset, hash.byteLength);
+  const seed = view.getUint32(0);
+  const prng = mulberry32(seed);
+
+  const values = new Array(dimensions);
+  for (let i = 0; i < dimensions; i++) {
+    values[i] = prng() * 2 - 1;
+  }
   return normalizeVector(values);
 };
 
@@ -43,7 +55,12 @@ const normalizeVector = (values: number[]) => {
     return values.map(() => 0);
   }
 
-  return values.map((value) => Number((value / magnitude).toFixed(8)));
+  const invMagnitude = 1 / magnitude;
+  return values.map((value) => {
+    const v = value * invMagnitude;
+    // Fast rounding to 8 decimal places
+    return Math.sign(v) * Math.round(Math.abs(v) * 1e8) / 1e8;
+  });
 };
 
 const toEmbeddingVectorRecord = (