lifedrawing.art
Applied AI & Automation Lab
Founded 2024 · Hackney, London · Non-profit weekly life drawing community | lifedrawing.art
overview
In 2024, I founded a non-profit life drawing community in London. While the public face is a traditional arts organisation, internally it operates as a rigorous production laboratory for applied AI, system automation, and decoupled architecture.
The community has real operational demands: ticketing, payment processing, marketing orchestration, and customer support. It is not a theoretical sandbox — it runs in production, has a paying user base, and system failures have immediate consequences.
This environment provides a commercial-risk-free proving ground to validate AI-first engineering patterns, migrate legacy workflows to automated pipelines, and test edge infrastructure under real-world loads.
architectural strategy
The platform is designed around strict service decoupling. Rather than relying on monolithic platforms or heavy CMS installations, the architecture wires together purpose-built services via clean APIs and orchestration layers. This creates natural system boundaries, allowing individual components or AI models to be swapped or upgraded without cascading risk.
infrastructure & orchestration
AI & automation (N8N)
- LLM customer support agent — Orchestrated an N8N workflow integrating a Gemini agent to manage inbound customer queries. The system is designed defensively: it resolves routine FAQ queries autonomously but is explicitly instructed to escalate ambiguous or complex issues to a human operator. A system that gracefully defers is significantly more valuable than one that confidently hallucinates.
- Marketing orchestration — N8N manages the complete social media pipeline, handling content scheduling, AI-assisted caption generation, and API dispatch. Token cost analysis on LLM calls drove practical implementations of prompt compression and response caching to maintain sustainable operational costs.
- Event lifecycle automation — Event publication, ticketing thresholds, and targeted marketing communications are orchestrated via event-driven triggers rather than manual intervention, demonstrating highly reliable "just-in-time" operational flows.
Self-hosted infrastructure & zero-trust networking
- Zero-trust tunnels — Deployed Cloudflare Tunnels to provide secure, authenticated access to self-hosted infrastructure. This ensures the N8N orchestration node and internal media servers (Immich) remain completely isolated from the public internet, accessible only via strict zero-trust rules while remaining fully reachable by webhook traffic.
- Media management — Self-hosted Immich instance running on local hardware to handle the session photography library, completely bypassing third-party storage fees while maintaining robust remote access.
Edge delivery & headless services
- Edge compute — Netlify Edge Functions handle request routing, API proxying, and content personalisation at the CDN layer, eliminating cold starts and reducing latency.
- JAMstack architecture — The public facing site is a statically generated Eleventy build, automatically deployed via CI/CD from Git. It requires zero server maintenance and offers maximum cacheability.
- Decoupled asset processing — ImageKit integration provides on-the-fly image transformation and modern format negotiation (WebP/AVIF), removing the need for batch image processing pipelines.
Operational data layer
- Headless state management — Leveraged the Google Sheets API as a lightweight, highly available database for operational metrics and scheduling, providing non-technical collaborators with a familiar interface while maintaining programmatic access for the N8N orchestrator.
applied engineering outcomes
Operating this stack in production validates theoretical AI patterns against actual operational friction. Key technical learnings include:
- Agent hygiene and guardrails — Proving that LLM integration requires robust fallback states and strict system prompts. Theoretical capabilities mean little if the agent cannot reliably identify when it needs human intervention.
- Automation observability — Complex orchestration pipelines accrue technical debt rapidly. Workflows must be designed with explicit logging, error-handling branches, and isolated execution scopes to remain maintainable.
- Token economics — Running LLMs in production forces a pragmatic approach to API usage. Efficient engineering requires treating token consumption as a first-class metric alongside latency and memory.
The project successfully demonstrates how modern AI orchestration, edge compute, and zero-trust networking can allow a single engineer to reliably operate infrastructure that previously required a dedicated team.