Skip to content
David Lonski

layer: services

Services.

Production systems and the agent work that taught me where it breaks.

status: active

01

MJL Flooring

live2026

Built and shipped the production website for MJL Flooring — established a customer inquiry pipeline for a small business that previously had no web presence, hardened the contact form against the spam overwhelming the owner, and improved local discoverability through a full SEO overhaul.

  • HTML/CSS/JS
  • Vercel Serverless
  • reCAPTCHA v3
  • Resend
View case study
02

Onsite OS

dev2026

Role-based AI access control for small teams — gives employees a safe way to run AI on company files, managers a way to oversee the work, and admins a way to audit the full trail. The non-enterprise version of what big AI platforms sell top-down.

  • AI
  • Role-based access
  • Audit trail
View case study

status: archived · agent infrastructure

01

OpenClaw Control Center

archived2026

Built a multi-agent orchestrator to learn where orchestration frameworks actually break — pushed it until it failed on purpose, then ditched the framework. The work product is the knowledge I build with now.

  • OpenClaw
  • Node.js
  • Ollama
View on GitHub
02

Pulse

archived2026

Built a local second brain to surface what mattered from calendar and mail without sending data to the cloud — proved that local LLMs could run real pipelines on bare metal.

  • Node.js
  • SQLite
  • Ollama
View on GitHub
03

PersonaModel

archived2026

Built a three-agent pipeline to turn mail and calendar into phone reminders — always-on, routing between local and API models to keep latency down and cost predictable.

  • Next.js
  • PostgreSQL
  • OpenClaw
View on GitHub
04

Second Brain Kit

archived2026

Turned years of chat history into a searchable vector vault to give agents memory they could actually use — embeddings, retrieval, and ground truth instead of hallucinated context.

  • Python
  • Embeddings
  • Vector search
View on GitHub
05

Local Lead Gen

archived2026

Built a swarm to run the full lead-gen pipeline end-to-end — scraped local businesses, generated sites, and ran outreach without a human in the loop. Collection, generation, and deployment as one automated flow.

  • Next.js
  • Playwright
  • n8n
View on GitHub

layer: stack

Stack

building
Cursor · Next.js · TypeScript · Node.js · Docker · Vercel
agentInfra
Ollama · MCP servers · local LLM routing · model gateways · context engineering · multi-agent pipelines · n8n
foundations
Python · SQL · PostgreSQL · pgvector · React · Git · GCP