AI Growth Engineer-AI Marketing Systems

The role LEC AI is building an AI-powered marketing operating system for LEC group's four brands (AI, robotics, beverages, UK-China trade), and productising it. You'll own it end-to-end: persona layers, digital humans, signal-triggered outbound, data indices, and the agent army. This is a builder role in live production, not an R\&D lab. What you'll build *Content \& persona systems* * Brand-brain content engine : RAG + knowledge-graph (pgvector, Neo4j) on a 4-layer memory stack, drafting on-brand content from real meeting transcripts across 4 brands, with per-brand compliance gates * AI persona layer : per-employee personas grounded in voice-ID'd meeting data: in-voice drafting to avatar video to conversational agents; Mandarin dubbing/localisation (one video, five languages) * Self-hosted digital-human pipeline : MuseTalk/LivePortrait/HeyGem-class OSS on RTX 4090-class GPUs; cloned TTS; RTMP streaming; sub-3s latency; AI livestream host for lead-gen and commerce * Compliant variation engine : one story, platform-native cuts per persona/brand, all disclosed and watermarked * Meeting→publish pipeline : WhisperX to diarization to clip to caption to multi-platform scheduling, fully automated *Growth \& revenue systems* * Signal-Triggered Growth Engine : Companies House, premises-licence, and venue-signal watchers firing persona-led outreach across our 16,681-venue LeadGen DB (Clay + n8n) * Proprietary Data Index Engine : quarterly indices from our operational data (on-trade, robotics deployment, UK-China trade), built for journalist and AI-engine citation * AI Fame Engine : newsjacking, journalist pitching, and SoS/Featured/Qwoted response automation on Donnie's insight memory * WhatsApp Trade-Ordering Agent : venue reorders via WhatsApp Business Platform (Twilio/360dialog) * AI phone agents : Vapi-class voice agents with UK data residency, consent-gated * Creative-Testing Factory : brand-brain generates 50-100 ad variants, auto-launches on Meta Advantage+/TikTok Smart+, auto-kills losers, winners feed back * Synthetic focus-group layer : pre-publish testing on AI personas built from first-party audience data * Agent-ready commerce : MCP/ACP endpoints, structured feeds, llms.txt/ai-catalog, schema for agentic discovery and buying The agent army ~40 agents spanning content, distribution, fame, revenue, analytics, commerce, and ops (LangGraph/CrewAI, self-hosted n8n as glue), including a Launch-Orchestrator ("product goes out, everything fires"), attribution tracking, consent/rights tracking, and brand-compliance checks (ASA/CAP/Portman, EU AI Act labelling). Must-have * 3+ years shipping production AI/LLM systems: RAG, embeddings, vector DBs, agent frameworks (LangGraph, CrewAI, or equivalent) * Hands-on with the 2026 gen stack: video-gen APIs (Veo/Kling/Seedance-class), voice cloning, avatar/lip-sync models * Self-hosting: Docker Compose, GPU inference, OSS model deployment (comfortable running Chinese OSS models locally) * Automation at scale: n8n, API integration across Meta, TikTok, LinkedIn, WhatsApp Business Platform, HubSpot, Companies House * Startup builder: shipped scrappy systems solo or in a tiny team Strong plus * Growth engineering: outbound infra, enrichment (Clay), deliverability, creative-testing loops * GEO/AI-search: how ChatGPT/Perplexity/AI Overviews cite and rank sources * UK compliance surface: GDPR biometric consent, PECR Reg 19, ASA/CAP; you build the gates * Mandarin or China-tech ecosystem familiarity (Douyin/WeChat/Xiaohongshu, ByteDance tooling) * Newsletter/community infra (beehiiv APIs), data-pipeline/BI experience for the Index Engine How we work * Join LEC AI's 6-engineer team, the team that built Donnie, our self-hosted AI Chief of Staff (4-layer memory, audio pipeline, MCP-connected) * Ship weekly; every build has a kill/scale threshold * Built to be productised for 10-100 person companies; you're building product, not rewriting internal tooling