Skip to content
makqai

CR8 AI — Agentic creative platform for global brands

I lead AI engineering for a hybrid SaaS and agentic AI platform that produces brand creative for Unilever, Philips and PepsiCo, from retrieval and agent workflows through to print-ready export.

Years
2024–present
Organisation
CR8 AI
Clients
Unilever · Philips · PepsiCo
Role
Head of AI Engineering
CR8 AI platform: generic pipeline patternA brief and brand assets feed retrieval, which grounds agent orchestration with tools. Generated work goes to human review and approval before print-ready export.Brief and brand assetsRetrievalbrand groundingAgent orchestrationwith toolsGenerationHuman review and approvalPrint-ready export
A brief and brand assets feed retrieval, which grounds agent orchestration with tools. Generated work goes to human review and approval before print-ready export.
  • Input
  • Processing step
  • Agent / AI step
  • Human review
  • Output

Context

CR8 AI (registered as ZOD Tech Ltd, London) runs a hybrid SaaS and agentic AI platform for global brands, including Unilever, Philips and PepsiCo. Engineering is based in Karachi. I joined in October 2024 as Head of AI Engineering.

What I lead

  • AI engineering for the platform: RAG pipelines, agentic workflows and LLM-integrated SaaS modules.
  • Approval and review systems for regulated content, so generated work passes human sign-off before it ships.
  • The AI roadmap across client workstreams: toolkit-based generation, chat interfaces and print-ready creative export.
  • A cross-functional team spanning AI, backend and platform engineering, which I build and lead.

How it works

At the pattern level, a request moves through a fixed sequence of stages. A brief and the brand’s own assets go in. Retrieval grounds the work in that brand material. An orchestration layer runs agents that act through tools rather than free text. Generated output goes to people for review and approval, and only approved work moves on to print-ready export.

Keeping each stage separate means a model, a tool or a review rule can change without touching the rest of the chain. Client-specific prompts, model choices and data models are not described here.

Stack

Python, FastAPI and Pydantic services; PyTorch; LLMs from OpenAI, Anthropic and Gemini alongside open-source models; RAG over vector databases; AWS (EC2, Lightsail) and Docker; React and React Three Fiber on the front end; Blender for 3D; and Adobe Illustrator automation through ExtendScript, UXP and CEP for print output.