Can I vibe code EasyAsk?

Build a custom, high-performance site search and merchandising microservice using Node.js (Fastify), Typesense, and React.

1. DATA MODEL & INDEXING:
- Create a Product schema in Typesense: id, title, description, sku, brand, categories (array), tags (array), in_stock (bool), price (float), vector_embedding (float array 1536-dim), created_at.
- Create a MerchandisingRule schema in PostgreSQL: id, trigger_query (string/pattern), pinned_skus (array), boosted_skus (array), buried_skus (array), start_date, end_date, active (bool).

2. CORE FUNCTIONALITY:
- GET /api/search: Accepts query string, page, limit, customer_group_id, and filters.
- If vector search is enabled, generate text embedding via OpenAI text-embedding-3-small and perform hybrid search in Typesense combining keyword score (BM25) and vector similarity.
- Check PostgreSQL for active MerchandisingRules matching the input query string.
- Re-rank search results by applying rule overrides: pin specified SKUs to exact positions, apply boost multipliers to boosted SKUs, and lower score/rank for buried SKUs.
- Dynamic Facet Generation: Return dynamic counts for categories, brand, in_stock, and price ranges along with the search hits.
- Autocomplete Endpoint (GET /api/suggest): Return fast prefix matching suggestions for products, categories, and top search queries.

3. MERCHANDISING ADMIN & SYNONYMS:
- Build a lightweight React/Tailwind admin dashboard to view search analytics (zero-result queries, top searched terms).
- Allow merchandisers to create, edit, and toggle boost/bury/pin rules per query term.
- Allow managing a custom synonym dictionary that gets synced into Typesense's synonym settings.

4. FAILURE MODES & OUT OF SCOPE:
- If OpenAI API times out, gracefully fallback to pure keyword search in Typesense without throwing a 500 error.
- Out of scope: Automated multi-currency real-time conversion during search indexing, or complex B2B contract price calculation inside the search microservice (receive customer group pricing pre-calculated).

$ each button prefixes agent-specific run instructions · build your own product, never copy proprietary code, trademarks or designs