Training · Knowledge delivery

A site guide and a starter kit that answers back.

The assistant in the bottom-right corner of this site. It knows every page, demo and program, adapts to the page you are on, and serves the AI 101 Starter Kit as a working assistant instead of a PDF.

Live on this site
DomainTraining and site guidance
StackClaude · Netlify Functions
ModelSwappable without a redeploy
Try itBottom-right corner
2Jobs: site guide and Starter Kit
5Page types it adapts to
0Keys in the browser: the model is called server-side
150Requests per visitor a day, then a polite stop
01
The problem

Training content dies the moment the PDF is downloaded.

Every AI 101 attendee gets a four-part Starter Kit: prompt patterns, a tool picker, a seven-day challenge and privacy rules. Useful material with a familiar fate: opened once, filed, forgotten.

A generic website chatbot is worse: it answers from the open web and contradicts what the training just taught. The brief was narrower and harder: make the kit itself answerable, and make the site navigable by asking.

Pain pointWhat it costWhat the build does
The kit goes unread
Practical guidance nobody retrieves when they need it.
The kit's content is the assistant's knowledge, one question away.
Generic answers
Off-the-shelf bots contradict the curriculum.
Answers come only from the site and the kit; anything else gets an honest redirect.
Visitors get lost
A big site, many pages, many demos.
It points to the right page or demo with one link.
Vendor lock and cost
A bot welded to one model and an open endpoint.
Model set on the server, per-visitor limits, and cached instructions.
02
The approach

Encode the knowledge. Keep the model replaceable.

The assistant's instructions are the site and the kit, written as rules: what each page offers, which demo fits which question, the tool-picker logic, the privacy thresholds. It never quotes consulting prices and never invents a fact.

The chatbot is the training material, running.

It is page-aware. On the enterprise page it leads with the gateway and security; on a demo page it explains that demo; on training pages it leads with levels and funding. Preset question pills change with the page.

The model is chosen on the server, so it can be swapped without touching the site. Each visitor has a daily limit, the instructions are cached so repeat messages cost a fraction, and the latest newsletter issue updates itself when a new one is published.

03
How it works

A thin system with a deliberate seam.

The knowledge and the model never fuse, so either can change without the other. Click a stage.

04
What it shows

What this work shows.

The smallest system that proves the whole method, and the one visitors use most.

Knowledge systems beat documents

The same content that goes unread as a PDF gets used as an answerable system.

Encode rules, not text

Faithful answers come from the material's logic, not from search over raw text.

Portability by design

The seam between knowledge and model is deliberate, in every system ResonAi builds.

Guardrails are part of the product

Limits, server-side keys and honest refusals make it safe to leave running.

05
See it working

Try something like it.

ClaudeNetlify FunctionsNetlify Blobs (limits)Prompt cachingServer-side model setting

What knowledge is sitting unread in your PDFs?

The same pattern fits onboarding guides, policy manuals and sales playbooks. Start with the document your team asks about most.

📅 Book a call