Case studies

Selected work.

Seven systems we built to ship as a solo developer. Each one solves a specific bottleneck — infrastructure, memory, content velocity, leadgen, agent coordination. Real numbers, real architecture, real lessons.

InfrastructureContent systemsSEO

VDL Factory — sixty-three sites, one operator

From zero to 63 owned websites in less than six months — production-grade infrastructure, automated deploys, ~8 hours per week of operational overhead.

Takeaway

Mono-repo + matrix CI + Cloudflare-first stack collapses the per-site overhead to near zero.

Read the case study →

AIEngineeringProductivity

Eleven agents — one terminal

A specialized council of eleven agents (data, infra, code, strategy, QA, UX, content, growth, security, platform, ops) coordinates work in real time from a single human operator.

Takeaway

Multi-agent role specialization replaces an engineering team for a solo operator's bandwidth profile.

Read the case study →

ProductAIOpen source

Hindsight — long-term memory for autonomous agents

Eleven specialized agents using the same memory layer across twelve months in production — 290+ persistent memories indexed and recalled across sessions.

Takeaway

When agents remember, the operator does not have to. That alone unlocks a 5–10× context multiplier.

Read the case study →

InfrastructureKubernetesFinOps

Cluster vdl-prod — Talos + ARC, scale-from-zero CI

Four CX33 Hetzner nodes in Helsinki running Talos Linux + Cilium + ARC ephemeral runners. Build farm spins up on demand, returns to zero in idle. ~35 €/month all-in.

Takeaway

A serious K8s control plane is now within reach of a one-person operation, not just a platform team.

Read the case study →

BackendConversionB2B

Pro-Leads — B2B insurance leadgen pipeline

Live pipeline serving French insurance brokers, built solo in six weeks. End-to-end: form intake, dedup, vertical pricing, broker distribution, transactional payouts.

Takeaway

Vertical leadgen as a productized engineering project, not a sales-led service business.

Read the case study →

MethodAIPractice

BMAD method — AI-assisted development workflow

A repeatable four-pass workflow (Brief, Model, Architect, Dev) that turns a one-paragraph idea into shippable code with the LLM in the loop end-to-end.

Takeaway

Method beats raw AI output. The framing of the four passes is what makes the model useful.

Read the case study →

ProductBackendDomains

DomSnap — drop-catching engine for premium domains

Daily ingest of AFNIC + Verisign zone files. Scoring pipeline filters thousands down to the single-digit shortlist actually worth bidding on. Auctions automated.

Takeaway

What used to take a domain broker now takes a cron job, a scoring rule set, and a watchlist.

Read the case study →