Kyle du Randt.
AI systems architect.
I build AI production infrastructure and agentic systems. I've been on the frontier since ChatGPT's first week — frontier models to local LLMs, training pipelines to production platforms. The marketing background is real and it changes what I build.
Technical depth and strategic context. Both sides of the same system.
Most builders understand the pipeline. Most strategists understand the output. Almost nobody understands both well enough to make one serve the other.
I've been the CMO responsible for what a pipeline produces — and I've built the pipeline. That combination means the system I design is already calibrated to the outcome before the first line of code runs.
On the frontier since week 1
ChatGPT launched, I was using it. Every major image model, video model, local LLM release — tracked, tested, and integrated into real production pipelines.
CMO background
Ran marketing across fintech, crypto, gaming, and consumer SaaS. Responsible for results, not just output. The difference shows in what I build.
Solo operator
Built ApeFX from scratch — platform, infrastructure, pipeline, brand. Live, real users, no team. The rig is the team.
Production at scale
55 live n8n workflows. 65 AI agents. Every system designed for unattended operation. Not prototypes — production.
Four startups. Three continents. The same pattern every time.
The businesses that won had the best infrastructure. Systems that compounded. Automation that ran while founders slept.
The ones that lost had tactics. Campaigns that reset every month. Agencies sending PDFs. No systems. Without systems, execution is just expense.
ChatGPT launches. I'm on it week one.
Not as a curiosity — as infrastructure. Started integrating generative AI into real production workflows immediately. Tracked every model release from that point forward.
CMO track. Four startups.
Ran marketing from zero across fintech, crypto, gaming, and consumer SaaS. Europe, North America, Africa. Responsible for results, not just strategy. Stopped hiring agencies because the machine I built worked better.
Built ApeFX from scratch.
Full-stack AI cinematic platform. Product, infrastructure, AI model integrations, mass automation pipeline. Everything solo. Live, real users. Every image and video on the platform produced by the system I built.
Fatboy Studios. The rig for everyone else.
55 live workflows, 65 agents, a vector context stack, and a client portal. The same infrastructure I built for ApeFX, deployed across engagements. I run it daily.
What the system actually produces.
55 workflows, not 55 people
Tools in daily production use.
Platform Engineering
- —Next.js 14+ (App Router, TypeScript)
- —Supabase (schema, RLS, migrations)
- —Vercel (deploys, edge, env)
- —REST APIs, webhook auth, rate limiting
AI Models — Image
- —Flux Kontext (BFL)
- —Nano Banana Pro (Gemini 3.1 Flash Image)
- —Seedream 4.x (ByteDance)
- —Recraft V4 (illustration + vector)
AI Models — Video
- —Kling 3.0 (Kuaishou — Omni + motion)
- —Seedance 2.0 (ByteDance direct)
- —Veo 3.1 (Google — 3 endpoints)
- —Higgsfield (Soul ID character)
Automation + Agents
- —n8n (self-hosted Docker fleet)
- —Claude API + MCP servers
- —Qdrant (vector context store)
- —Context API (per-client /context endpoint)
- —Ollama + Qwen (local LLM tier)
LLMs + Models
- —Claude Sonnet/Opus (complex reasoning)
- —Claude Haiku (high-volume routine)
- —Qwen 3.5 local (execution tier)
- —GPT Image 2 (character generation)
- —NB2 (thinking_budget 1024 production)
Infra + Observability
- —Docker (n8n, Qdrant, Grafana, Ollama)
- —Cloudflare (DNS, R2, Workers)
- —Sentry (Next.js error tracking)
- —Grafana + Prometheus (local dashboards)
- —Uptime Kuma (15+ service monitoring)
The team is the machine.
The machine doesn't need managing.
Fatboy Studios