// Semantic Search — Cosine Similarity Demo // This exercise uses a pre-computed embedding space (simplified) // Real RAG uses an embedding model API + vector database // Simplified 8-dimensional embedding vectors for demo documents const docs = [ { id: 1, text: "SQL window functions allow calculations across partitions of rows", vec: [0.9, 0.1, 0.0, 0.2, 0.0, 0.0, 0.1, 0.0] }, { id: 2, text: "Transformers use self-attention to process sequences in parallel", vec: [0.0, 0.0, 0.9, 0.1, 0.0, 0.2, 0.0, 0.0] }, { id: 3, text: "RAG combines retrieval with generation to answer questions from documents", vec: [0.0, 0.0, 0.1, 0.9, 0.1, 0.2, 0.0, 0.0] }, { id: 4, text: "dbt transforms data in the warehouse using SQL SELECT statements", vec: [0.8, 0.2, 0.0, 0.0, 0.9, 0.0, 0.0, 0.0] }, { id: 5, text: "Vector databases store embeddings for fast similarity search", vec: [0.0, 0.0, 0.2, 0.8, 0.0, 0.9, 0.0, 0.0] }, { id: 6, text: "Airflow orchestrates data pipelines as directed acyclic graphs", vec: [0.7, 0.9, 0.0, 0.0, 0.3, 0.0, 0.0, 0.0] }, { id: 7, text: "Fine-tuning adapts a pre-trained LLM to a specific task", vec: [0.0, 0.0, 0.6, 0.2, 0.0, 0.0, 0.9, 0.0] }, { id: 8, text: "Snowflake uses virtual warehouses to separate compute from storage", vec: [0.6, 0.1, 0.0, 0.0, 0.8, 0.0, 0.0, 0.9] }, ]; // Query embeddings (simplified, 8-dimensional) const queries: Record = { "how does attention work in neural networks": [0.0, 0.0, 0.9, 0.2, 0.0, 0.1, 0.0, 0.0], "what is RAG and vector search": [0.0, 0.0, 0.2, 0.9, 0.0, 0.8, 0.0, 0.0], "sql analytics and data transformation": [0.9, 0.1, 0.0, 0.0, 0.8, 0.0, 0.0, 0.2], "pipeline orchestration and scheduling": [0.5, 0.9, 0.0, 0.0, 0.2, 0.0, 0.0, 0.0], }; // ── YOUR TASK ──────────────────────────────────────────────────────────────── // Implement cosineSimilarity(a, b) and search(queryVec, topK) // Then run a demo to see which docs match each query function cosineSimilarity(a: number[], b: number[]): number { // TODO: compute dot product / (|a| * |b|) // Hint: dot product = sum of a[i] * b[i] // Magnitude = sqrt(sum of x[i]^2) throw new Error("Not implemented — fill this in!"); } function search(queryVec: number[], topK: number = 3): typeof docs { // TODO: score each doc, sort by similarity descending, return top K throw new Error("Not implemented — fill this in!"); } // Demo — run this after implementing above function runDemo() { const output = document.getElementById("output")!; let html = ""; for (const [queryText, queryVec] of Object.entries(queries)) { html += `

Query: "${queryText}"

${sim} — ${doc.text}
${(e as Error).message}

"; } output.innerHTML = html; } document.getElementById("runBtn")!.addEventListener("click", runDemo);