Natural Language Search Configurator

AI-assisted interface that transforms free-text interior designer queries into structured Algolia search JSON using OpenAI’s Structured Outputs and Zod validation.

AI / Search Engineeringapp.rugandkilim.com

I. The Challenge

Rug & Kilim’s team wanted search to mirror how designers actually think—phrases like “6×9 Scandinavian under $5000 in beige” rather than dropdown filters. The challenge was to interpret that natural input, convert it into valid Algolia queries, and still respect business constraints such as product taxonomies, stock status, and pricing logic — all while remaining editable by non-technical teammates.

II. The Solution & Architecture

Built a modular Natural Language → Algolia JSON pipeline using OpenAI’s responses.parse() with Zod schema enforcement. Added a Neon (Postgres) backend for storing evolving prompt rules and a collaborative “Questions Bank” where staff could log and refine real-world queries. The frontend provides authenticated editors an interface to test, adjust, and validate prompt behavior in real time.

System Architecture

Next.js Interface

Authenticated playground to test AI-to-Algolia translations

OpenAI Structured Outputs

Converts natural text into Zod-validated JSON schema

↓

Zod Schema Validation

Guarantees strict typing of facet and numeric filters

Algolia Index (wp_posts_product)

Executes generated structured queries instantly

↓

Rules API (Neon DB)

Stores custom prompt modifiers and system rules

Questions Bank

Centralized log of 150+ designer search phrases

↓

WordPress Auth Layer

Restricts access to internal team members only

Netlify + Cloudflare

Serverless hosting and edge caching for AI queries

III. Technology Stack

Frontend

Next.js 15 + App Router
Tailwind UI + ShadCN Components
React Hooks for Rules & Questions
JWT Auth via WordPress API

Backend & AI

OpenAI Responses API (Structured Outputs)
Zod Schema Validation
Algolia Search API
Next.js Server Actions

Database & Automation

Neon (Postgres Serverless)
Prompt Rules & Questions Bank Tables
Automated Rule Activation Flow
Zod Error Logging for Debugging

Infrastructure & Analytics

Netlify Deployments + Edge Functions
Cloudflare Cache + Analytics
Algolia Logs for Search Accuracy
PostHog Behavior Tracking

IV. Results & Impact

150 +

Real designer queries collected and parsed

100 %

Schema-validated Algolia JSON with zero malformed responses

< 6 s

Average AI + Algolia end-to-end latency

Live Beta

Used internally to train AI prompt rules and monitor accuracy

Key Learnings

•Combining Zod schemas with OpenAI Structured Outputs ensured consistent, production-ready JSON.
•Allowing non-engineers to add rules and queries accelerated prompt training and context coverage.
•Schema-driven search bridges AI interpretation with deterministic E-commerce filters.