← all work
Production · SaaS · 2026 SHIPPED

Multi-model AI Content Generator.

Multi-model AI content generator with citation retrieval, deployed across Cloudflare Workers (API) and Pages (frontend).

Cloudflare Workers Cloudflare Pages Vite OpenAI GPT-5 AI-Search
Client undisclosed · NDA · technical surface only

Problem

Content generation that’s actually useful needs (a) multiple models compared head-to-head, not a single black-box choice, and (b) citations grounded in retrieval, not hallucinated.

Approach

Worker API on Cloudflare with a curated model list endpoint (so the frontend never has to ship a stale enum). The frontend (Vite + Pages) calls the same /score and /propose endpoints with models[] and count parameters, getting parallel multi-model output back.

Citations come from AI-Search as the primary source, with the worker handling fan-out, dedup, and response shaping.

What I shipped

  • Multi-model proposal + scoring across N models with per-call counts
  • Curated model list endpoint (so the FE doesn’t drift from the BE)
  • AI-Search citation retrieval pipeline as the primary grounding source
  • CORS hardening (ALLOWED_ORIGINS Worker secret) after a production-blocking misconfiguration
  • Bug fix for GPT-5 returning empty content on certain prompt shapes
  • Stale-page bug fix on the production frontend
⚜   highlights   ⚜
  • Multi-model proposal + scoring; curated model list endpoint
  • AI-Search as primary citation source
  • Hardened CORS, fixed empty-content GPT-5 edge case
← all engagements