Autonomous AI Sales Agent — Knowledge Base & Dynamic Bundling

Dynamic Cart Generation — Rules Engine

The agent does not freestyle bundles. It runs every cart through 4 rule families, in order. The rationale column tells the founder why each rule exists — most map directly to known student behavior patterns from past calls.

Live Cart Simulator

Add services to a mock cart. The right panel shows exactly which cross-sell / upsell / package / discount rules fire — this is what the AI agent will pitch.

AI Agent Grounding

Auto-generated from the knowledge base above. Paste this into the AI agent's system prompt or load the JSON at runtime. Regenerates every time you edit a service or rule.

System Architecture

Same deployment pattern as the Ops Graph artifact: static HTML in GitHub → auto-deployed via Cloudflare Pages → fetched by the AI agent and rendered by ops at kb.nbyula.com.

Step-by-step deployment (GitHub + Cloudflare Pages)

1. 1
   Create the GitHub repo

   Repo name: nbyula-sales-kb (private). Add this HTML as index.html at root. Also commit an empty config.json — this will be the retrieval JSON the AI agent fetches.

   gh repo create nbyula/nbyula-sales-kb --private --add-readme
   git clone git@github.com:nbyula/nbyula-sales-kb.git
   cd nbyula-sales-kb
   # drop this HTML in as index.html
   git add . && git commit -m "initial KB v1" && git push

2. 2
   Connect to Cloudflare Pages

   In the Cloudflare dashboard → Workers & Pages → Create → Pages → "Connect to Git". Pick the repo. Build settings:

   • Framework preset: None (static HTML)
   • Build command: leave empty
   • Build output directory: /

   Hit Deploy. Cloudflare gives you a *.pages.dev URL within ~30 seconds.

3. 3
   Map the custom domain

   Pages project → Custom domains → Add → kb.nbyula.com. Cloudflare auto-creates the CNAME. SSL auto-issued in ~1 minute.

4. 4
   Wire the AI agent to fetch /config.json

   In the AI agent platform (Vapi / Bland / our own), add a tool / function call at session start:

   async function loadKB() {
     const r = await fetch('https://kb.nbyula.com/config.json', {
       headers: { 'Cache-Control': 'no-cache' }
     });
     return await r.json();
   }
   // inject the result into the system prompt as <services> and <rules> blocks

   Use the AI Agent Grounding tab to copy the exact JSON / prompt template.

5. 5
   The daily edit loop

   1. Open https://kb.nbyula.com
   2. Edit a service price / add a rule (everything is inline-editable)
   3. Click Export JSON → save as config.json
   4. Commit + push to GitHub (or use the GitHub web UI to replace the file)
   5. Cloudflare auto-deploys in 30 sec → next AI call picks up the new prices

   Optional follow-up: add a Cloudflare Worker that lets ops PATCH /config.json directly from the page — eliminates the git commit step entirely.

6. 6
   Resource ask (per sprint plan)

   One dedicated engineer to handle the AI agent's runtime fetch + the Cloudflare Worker for in-place edits. Estimated effort: 3–5 engineering days total once this HTML is in place.

Why GitHub + Cloudflare (and not a database)