EthosPowerAIDelivery
From a single extraction layer to a whole self-hosted platform, built on open foundations, on your infrastructure, and yours to keep rather than rent. Every build makes a named person faster or better informed; none replaces one.
Automations
Stop doing it by hand
Repetitive, manual work between systems that don't talk to each other — tool-to-tool sync, document workflows, approval and handoff chains, data pipelines, notifications, and bridges to legacy systems.
What this looks like in practice
A field engineer's battery discharge-test reports — voltage, temperature, and discharge current across up to 54 battery blocks — captured by voice through a privacy-first AI agent running on the company's own server.
AI Agents
A tireless digital teammate
Customer support agents, data processing agents, workflow agents that orchestrate multi-step processes, knowledge-base agents, research assistants, and drafting or triage agents — grounded in your data, not a generic chatbot. Built on LangChain/LangGraph, Claude, OpenAI, or Ollama, with Qdrant or ChromaDB for retrieval.
What this looks like in practice
An energy-sector company was losing leads to manual follow-up. I built a multi-touchpoint capture system that checks the ERP/CRM for an existing match on email, company, or phone before scoring a lead — deduping merges into the original record, while genuinely new contacts trigger a three-email welcome sequence behind human approval gates.
Websites & Web-Apps
Your public presence, or a tool your team logs into
Marketing websites with built-in AI — semantic search, retrieval assistants, automated content pipelines, edge deployment (this site runs on the same stack, and I'll walk you through it) — and internal tools: dashboards, customer and partner portals, booking and intake apps, data-entry and approval apps, AI-assisted apps. Built on Next.js 15, Payload CMS, and Cloudflare Workers. You own the code.
What this looks like in practice
eBatt.ai grew from spreadsheets, through Django web apps, into a live battery-quoting engine: IEEE 485 sizing and EN 50272-2 venting free, selection, RFQ and checkout behind them. It puts the methods I have used on standby DC power since 1992 into software, and I built it myself on a current, production-grade stack.
MCP Connectors
Make your systems answerable from your customers' own AI tools
Your customers increasingly ask an AI assistant before they ask you. A Model Context Protocol connector — the standard the major AI platforms now support — lets their assistant query your systems directly and get an authoritative answer, instead of guessing from whatever it scraped. An API your customers' tools can actually reach, with the access rules and audit trail you set.
I build these in production. This site runs one at /api/mcp, exposing its engineering knowledge base to any AI client that speaks the protocol — so you can point your own assistant at it and see exactly how the pattern behaves before commissioning your own.
Platforms
The whole system, end-to-end
Data capture, an AI processing layer, a knowledge store, operational records, automation and follow-up, and self-hosted infrastructure — architected and built as one system, not a collection of point tools. Scoped to your system, discovery-first.
What this looks like in practice
An engineering consultancy held decades of reports and design documents, and wanted its people to search that archive with AI — without exposing proprietary knowledge to cloud models or anyone outside the firm. I built a self-hosted RAG knowledge base with enterprise document management, end-to-end encryption, and complete data sovereignty.
All prices in USD (converted from NZD), exclusive of GST. Advisory rate: $129/hr.
Not sure which of these fits?
Every engagement starts with a conversation, not a quote. I'll tell you honestly if a smaller automation solves it before I pitch you a platform.