BALANCED+ Inc. · 2026

Bringing AI into a client's operation, on their own infrastructure

I pitched, designed, and built an internal AI assistant with a retrieval knowledge base over roughly 50 documented processes, hosted on the client's private network rather than a vendor's, so their operational knowledge never leaves the building.

Role

Pitched it, designed it, built it, and own it

Focus

AI implementation, Claude API, RAG, Internal tooling

The problem

The process documentation I was producing was accurate but sat in Word files on a network drive. Nobody on a production floor is going to open a fifty-document folder to answer a question mid-shift, so the documentation was correct and unused, which is the same as not having it.

The obvious fix is a chatbot over the documents. The obstacle is that the documents describe how a business actually runs, and the client was not going to accept that being uploaded to a third-party service.

What I did

I pitched the concept to the client stakeholder, got approval, and built it end to end rather than handing it to a vendor. It uses the Claude API for the language model but runs on an internal Linux server inside the client's private network, reachable in a browser on an internal domain, with DNS and hosting coordinated with the registrar.

The retrieval knowledge base is built from the process documents produced by the analysis project, so the two efforts compound: every process I document makes the assistant more useful, and the assistant is the reason the documentation gets read at all.

Separately I integrated a Microsoft 365 MCP server for a client stakeholder, and I use Claude Code daily as an automation and tooling driver, this site included.

Why it was the right call

The decision that mattered was not the model. It was choosing self-hosted over SaaS, which cost more effort up front and is the only reason the client agreed to put real operational knowledge into it.