IbuildAIsystemsthatworkwhileyousleep.
I design and deploy AI agents, n8n + Make automations, and Bubble SaaS systems for operators who want leverage.
Live systems. Built, deployed, running.
A selection of production systems I've designed and shipped. Tap any to open the case study.
From idea to running system — in four moves.
Every system I ship goes through the same four-stage pipeline. No mystery, no theatrics.
IDEA
Capture the messy real-world problem before optimizing anything.
- Frame the operator's job to be done
- Sketch the smallest end-to-end path
- Pick the killer constraint that defines done
ARCHITECTURE
Draft the map. Choose primitives that won't betray you at scale.
- Pick orchestration engine (n8n / Make / custom)
- Decide datastore + memory + queues
- Define observability + retry budgets
AUTOMATION
Wire the nodes. Push events. Watch the system come alive.
- Build node graph from triggers → outputs
- Bake in idempotency + replay safety
- Ship the smallest live cut, then iterate
SCALE
Tune for cost, latency, and resilience. Watch the dashboard flatten.
- Profile bottlenecks, parallelize hot paths
- Cache the expensive thinking
- Hand off to the operator with a runbook
Build logs, teardowns, and unedited live builds.
The same lab, on tape. New drops weekly.
Building a 12-agent mesh in n8n + LangGraph
Building a 12-agent mesh in n8n + LangGraph
From a blank canvas to a running agent mesh that resolves real support tickets. Live, no cuts.
n8n vs. Make: which one survives at 2M events / day?
I ran both engines at scale for two weeks. Here's where each one breaks, and why I still use both.
Stripe metered billing in Bubble in 14 minutes
The pattern I use on every SaaS build. Includes the workflow JSON in the description.
Shipping a voice receptionist in one weekend
Weekend timeline of building, deploying, and calling my own receptionist agent.
An AI marketplace that prices itself dynamically
Pricing as a product feature. Embeddings + simulated demand to find equilibrium.
Agent memory: why your bot forgets and how to fix it
Per-agent shards, summarization passes, and the eviction policy that finally worked.
Cold pipeline 5x in 30 days — full autopilot
The exact graph, scoring rules, and Slack handoff that drove the lift.
My 2026 no-code stack (the one I actually use)
What stayed, what I dropped, and what I'd never use in production.
Currently building. In the open.
Half-formed prototypes that earn their way into the work above.
Self-healing workflows
Workflows that detect their own failure modes, ask an LLM to diagnose, and patch themselves before paging me.
Agent simulator
A sandbox that replays real user traces against new agent prompts before deploy. Treats prompts like code.
Voice → workflow
Speak the workflow you want. The system drafts a runnable n8n graph and asks for confirmation.
Marketplace pricing oracle
An always-on agent that watches my AI marketplace, simulates demand, and proposes price changes hourly.
Let's build something serious.
Tell me what you'd like to operationalize. I'll respond within 24 hours.