Summary

Addresses #5478 (point 3: recommendation engine template).

• Adds a new recommendation template to feast init demonstrating product recommendations using vector similarity search
• Uses pre-computed embeddings via all-MiniLM-L6-v2 (384 dimensions) with SQLite online store (vector_enabled: true)
• Includes 12 sample products across 4 categories (Electronics, Books, Sports, Home)
• Demonstrates retrieve_online_documents_v2 API for top-k similarity search

New files

• templates/recommendation/bootstrap.py — generates product catalog with embeddings
• templates/recommendation/feature_repo/feature_definitions.py — Entity, FeatureView with vector_index, FeatureService
• templates/recommendation/feature_repo/feature_store.yaml — SQLite config with vector search
• templates/recommendation/feature_repo/test_workflow.py — end-to-end demo: apply, materialize, search

Modified files

• sdk/python/feast/cli/cli.py — register recommendation in template choices (+1 line)

Test plan

• feast init test_rec --template recommendation creates project successfully
• feast apply registers entity, feature view, feature service
• python test_workflow.py runs full cycle: apply → materialize → vector search → teardown
• Vector search returns relevant results (e.g. "gaming laptop accessories" → electronics items)
• Idempotent feast apply (second run reports no changes)
• Existing local template unaffected
• ruff check and ruff format pass