Massachusetts Building Code — n8n RAG Assistant

Why This Stack?

Reading dense code sections against a deadline is slow and error-prone. A retrieval-augmented workflow keeps answers grounded in the chunks you indexed (current code cycle, amendments, and your annotations) instead of the model guessing from memory. n8n wires ingestion, retrieval, and the LLM in one place you can audit, version, and run on your own infrastructure.

The Workflow

Step 1 — Curate and ingest code sources

Collect the official Massachusetts Building Code sources you are licensed or permitted to use (current edition, local amendments, and AHJ bulletins). In n8n, use nodes to fetch or read files, split text into chunks, generate embeddings, and upsert into your vector database (e.g. Postgres + pgvector, Pinecone, or Qdrant). Schedule re-ingestion when the code cycle updates.

Step 2 — Build the RAG chat workflow in n8n

Create a workflow triggered by a Webhook node. On each request, accept message and optional sessionId (UUID) so downstream nodes can load prior turns from memory or a thread store. Run similarity search on the user query, assemble a prompt with citations or chunk IDs, call your LLM provider, and return a JSON body with a single assistant string field (e.g. reply).

Step 3 — Try the public lab (optional demo)

We host a small Next.js page that proxies chat to your n8n webhook so the URL and tokens stay server-side. Open the Building Code lab to send test questions. The lab generates a session id on first send and sends it on every follow-up so your workflow can thread context without storing full history in the browser.

Limitations

• Not legal or engineering advice. Outputs depend on your indexed material and model settings; always verify against the official code and your Authority Having Jurisdiction (AHJ).
• Demo availability depends on configuring N8N_MASS_BUILDING_WEBHOOK_URL in the deployment environment.

Estimated Monthly Cost

Total depends entirely on hosting and model choices.