Can I vibe code Nosto?

Build a light-weight Product Recommendation & Merchandising Engine for a Shopify store using Next.js (App Router), Supabase (PostgreSQL with pgvector), Cloudflare Workers, and Upstash Redis.

CORE ARCHITECTURE & ENTITIES:
1. Data Schema:
   - products: id, shopify_id, title, handle, price, inventory_quantity, tags (array), vector_embedding (vector(1536)).
   - events: id, session_id, user_id, event_type (view_product, add_to_cart, purchase), product_id, timestamp.
   - merchandising_rules: id, collection_id, pinned_product_ids (array), boosted_tags (array), buried_tags (array), hide_out_of_stock (boolean).
   - recommendations_cache: key (product_id + strategy), recommended_ids (array), updated_at.

2. Ingestion & Event Pipeline:
   - Create a fast API route `/api/v1/event` running on Cloudflare Workers / Vercel Edge.
   - Accept anonymous `session_id` and log interaction events (`view_product`, `add_to_cart`, `purchase`) directly into an Upstash Redis Stream for asynchronous ingestion into Supabase.
   - Include a Shopify Webhook handler (`/api/webhooks/shopify`) for `products/create`, `products/update`, and `orders/create` to continuously update product catalog, stock state, and co-occurrence graphs.

3. Recommendation Engine Logic:
   - Strategy 1: "Frequently Bought Together" — Calculate item co-occurrence matrices from the `events` table for orders containing multiple product IDs. Cache top 4 related items in Redis.
   - Strategy 2: "Visually / Semantically Similar" — Compute cosine similarity using OpenAI embeddings or pgvector over `products.vector_embedding`.
   - Fallback Logic: If co-occurrence or vector similarity returns <4 items, backfill results using top-selling products in the same category. Exclude currently viewed product and out-of-stock items if `hide_out_of_stock` is true.

4. Merchandising Rules Engine:
   - Build an API endpoint `/api/v1/recommendations` accepting `product_id`, `strategy`, and `collection_id`.
   - Apply `merchandising_rules`: Force `pinned_product_ids` to positions 1..N, multiply scoring weight for `boosted_tags` by 1.5x, penalize `buried_tags` by 0.5x, and strip out-of-stock products when enabled.

5. Front-End Storefront Widget:
   - Build an ultra-lightweight client-side Web Component or React widget (<10KB) that embeds directly on Shopify PDPs.
   - Fetch recommendations asynchronously using `fetch()` with sub-50ms response budget, rendering product card image, title, price, and instant "Add to Cart" form post.

OUT OF SCOPE:
- Full search UI, visual drag-and-drop merchandising editor, dynamic pop-ups, shoppable UGC, email personalization triggers, and multi-currency edge localization.

FAILURE MODES TO HANDLE:
- Redis memory exhaustion during extreme traffic spikes (fallback directly to raw SQL query on Supabase with 1-second timeout).
- Cold-start problem for newly created products with zero event history (default to semantic vector similarity).
- Stale stock displaying in widgets (check live Redis stock count key before rendering response payload).

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