SearchAF Logosearchaf
← Back to Guides
Advanced

Implementing Semantic Search

Leverage AI-powered vector search to power intelligent answer generation

Semantic search uses AI to understand the meaning and intent behind search queries, delivering more relevant results than traditional keyword matching.

Traditional search matches keywords exactly:

  • Query: "running shoes" → Only finds products with those exact words

Semantic search understands meaning:

  • Query: "footwear for jogging" → Finds running shoes, sneakers, athletic shoes
  • Query: "workout gear for feet" → Also finds relevant products

How It Works#

SearchAF's semantic search uses:

  1. Embedding Models: Convert text to vector representations
  2. Vector Database: Store and search high-dimensional vectors
  3. Similarity Search: Find products with similar meaning
  4. Hybrid Ranking: Combine semantic and keyword signals

Via Dashboard#

  1. Navigate to your project settings
  2. Go to "Search Configuration"
  3. Toggle "Semantic Search" on
  4. Choose your embedding model:
    • Fast: Lower latency, good accuracy
    • Accurate: Higher latency, best accuracy
  5. Click "Save"

Via API#

Enable semantic search in your search requests:

curl https://searchaf-api.antfly.io/api/v1/search \
  -H "X-API-Key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "comfortable shoes for hiking",
    "semantic": true,
    "hybrid_weight": 0.7
  }'

The hybrid_weight parameter controls the balance:

  • 0.0: Pure keyword search
  • 0.5: Equal weight to semantic and keyword
  • 1.0: Pure semantic search

Best Practices#

Optimize Product Descriptions#

Semantic search works best with rich, descriptive content:

Good:

Women's Trail Running Shoes - Waterproof hiking footwear with
superior grip and cushioning for outdoor adventures

Bad:

Shoes - Blue - SKU12345

Use Natural Language#

Write product descriptions the way customers search:

  • Include synonyms and related terms
  • Describe use cases and benefits
  • Add context about who it's for

Configure Relevance Tuning#

Fine-tune which fields contribute to semantic matching:

{
  "semantic_fields": {
    "title": 0.4,
    "description": 0.4,
    "tags": 0.2
  }
}

Advanced Features#

Multi-Language Support#

Semantic search works across languages:

{
  "query": "chaussures de course",
  "language": "fr",
  "semantic": true
}

Custom Embeddings#

Use your own embedding model:

  1. Train or select a custom model
  2. Upload to SearchAF
  3. Configure in project settings

Query Expansion#

Automatically expand queries with related terms:

{
  "query": "laptop",
  "semantic": true,
  "expand_query": true
}

This might expand to: "laptop computer notebook portable pc"

Monitoring Performance#

Track semantic search effectiveness:

  1. Click-Through Rate: Are users clicking results?
  2. Conversion Rate: Are searches leading to purchases?
  3. No-Result Queries: Which queries return nothing?
  4. Latency: Is response time acceptable?

Cost Considerations#

Semantic search uses more compute resources:

  • Embedding Generation: ~2-5ms per query
  • Vector Search: ~10-50ms depending on catalog size
  • Storage: ~1KB per product for embeddings

Plan your tier accordingly based on query volume.

Next Steps#