Rug & Kilim's team wanted search to mirror how designers actually think, phrases like "6x9 Scandinavian under $5000 in beige", rather than wrestling with a wall of dropdown filters.
02The Solution
+
A middleware layer turns free-text into precise Algolia queries: OpenAI structured outputs, constrained by a Zod schema, translate a designer's phrase into a validated query object mapping size, material, color, price, and stock status, all while staying editable by non-technical teammates.
03Features
+
Designers type how they think, e.g. '6x9 Scandinavian under $5000 in beige', instead of using filters.
OpenAI structured outputs translate the phrase into a precise Algolia query object.
Zod constrains the model's output so only valid, well-typed queries reach Algolia.
Maps language to real attributes: size, material, color, price, and stock status.
Built so non-technical teammates can adjust the mapping logic.
Reliably delivers precise results without hand-built filter UIs.
04System Architecture
+
Input
Free-text query
API
Next.js API
AI
OpenAI + Zod
Search
Algolia
Output
Ranked results
Press enter or space to select a node. You can then use the arrow keys to move the node around. Press delete to remove it and escape to cancel.
Press enter or space to select an edge. You can then press delete to remove it or escape to cancel.
Input
Free-text query
API
Next.js API
AI
OpenAI + Zod
Search
Algolia
Output
Ranked results
05Technology Stack
+
Next.js · TypeScript
OpenAI Structured Outputs · Zod
Algolia · Neon (Postgres)
06Key Learnings
+
01Constraining the model with Zod turned free-form AI output into reliable, well-typed Algolia queries.
02Mapping natural language to real product attributes mattered more than the prompt, the schema was the contract.
03Tuning against 150+ real-world queries surfaced the edge cases that made it production-ready.
nahom endale