JOY Automart's autonomous Chief Marketing Officer. ZIFA takes JOYADH's strategic command, queries ONI for the right customer profile, and generates pixel-perfect ad copy, image prompts, and budget allocations — all in valid JSON. No hallucinated parts. No vague tone. Every asset bound to a real SKU.
Hyper-personalise every asset. Tailor a DIY hobbyist ad differently from a B2B wholesale mechanic ad. Bundle relevant parts to lift AOV. Never invent a feature, spec, or fitment that ONI has not verified.
5 ad variants per campaign · platform-tuned (Facebook · Bikroy · Google · YouTube).
Native Bengali tone for local hobbyists · operator English for B2B wholesale mechanics.
Every ad variant ships with a visual_prompt_for_image_gen ready for Nano Banana / GPT Image 1.
Recommends spend split across channels given JOYADH's envelope + ONI's CAC history.
Brake-pad + rotor + sensor bundles tuned to AOV uplift signals from ONI.
Every generated variant ships with a one-click 'Copy as Ads JSON' button for Meta + Google Ads.
ZIFA cannot hallucinate a part. Every ad variant references a real SKU from the catalog. Every fitment claim is verified against JOY VIN. Marketing without the usual creative drift.
ZIFA drafts. The launch button still lives in the marketing team's Meta / Google / Bikroy console. By design — the founder retains the final 'yes' on every spend.
| Layer | Tech | Purpose |
|---|---|---|
| Brain | Claude Opus 4.8 + persona prompts | Tone-controlled creative + JSON-strict output |
| Schema | Strict-JSON campaign blueprints | ad_variants[] · budget_allocation_request · projected_roi |
| Memory | db.zifa_sessions | Multi-turn brand consistency |
| Visual prompts | Nano Banana / GPT Image 1 ready | Image-gen wire on each variant |