Internal support bot for an architecture firm
A 60-person architecture firm wanted an internal chatbot trained on their building-code reference library, without sending any of the documents to a hosted LLM.
Junior architects were spending hours searching across PDFs of building codes, specification sheets, and the firm's own design standards. The partners were not comfortable uploading any of those documents to a hosted AI service — much of it is regulated reference material and the rest is competitive intellectual property.
We deployed Ollama with Llama 3.1 8B on a dedicated workstation already running in the firm's office, set up a Qdrant vector store, and built an n8n ingestion pipeline that watches a Dropbox folder for new specifications. A custom OpenWebUI install gives every architect a chat interface that retrieves the right code passages with citations.
The team replaced an average of 90 minutes of daily code-searching with a chat that answers in seconds. The firm kept every document on its own infrastructure and avoided a six-figure annual subscription to a hosted research tool.
