The chat you had on our homepage is a real AI system built by our team. Here’s exactly how it works — and why it matters for your business.
Why we built it
Most business websites use static forms or multiple-choice quizzes to figure out what a visitor needs. We’re an AI engineering company — so we built an AI that actually talks to you.
The assistant asks a few questions about your business, listens to your answers, and recommends the right service based on what you actually said — not which checkbox you clicked.
We designed, built, tested, and iterated on this in less than a day. That’s the point: AI makes things like this fast and cheap to create. The same kind of automation can be applied to your business operations.
From your first message to a personalized recommendation in under 2 minutes
💬
Your Browser
You type a message about your business. The chat sends it to our server — your conversation stays in your browser until you’re done.
your message
🛡️
Netlify Edge Function
Our server-side gatekeeper. Runs close to you for speed and handles security before anything reaches the AI.
rate limitinginput validationprompt protection
sanitized request
🧠
Claude Sonnet 4.6
Anthropic’s AI reads your conversation, understands your business context, and generates a personalized recommendation. Only real services at real prices — no hallucination.
your data is never used for training
streaming tokens
⚡
Real-Time Streaming
The response streams back word-by-word to your browser. You see the answer appear as it’s written — no waiting.
📧
Team Notified
Our team gets an instant email with your recommendation — so if you book a call, we already have context.
🔄
Automatic Fallback
If anything goes wrong, the site switches to a static quiz. You always get a recommendation.
$0.015
per conversation — less than two cents
The tech stack
Claude Sonnet 4.6 by Anthropic
The AI brain. Claude reads your messages, understands the context of your business problem, and generates a personalized recommendation. It’s the same AI that powers tools used by companies like Notion, DuckDuckGo, and Quora.
Netlify Edge Functions server infrastructure
The middleman between your browser and Claude. It runs at the network edge (close to you, wherever you are), keeps our API keys secure, enforces rate limits, and streams Claude’s response to you in real time — that’s why you see the text appear word by word.
Streaming responses real-time UX
Instead of waiting 3–5 seconds for a complete response, you see the AI’s answer appear as it thinks. This uses Server-Sent Events (SSE) — the same technology that makes ChatGPT feel responsive. It’s a small detail that makes the experience feel alive.
Prompt engineering the invisible layer
Behind every AI conversation is a carefully written set of instructions that the AI follows. Ours tells Claude exactly what services we offer, what prices and timelines are correct, how to stay on topic, and when to make a recommendation. This is what prevents the AI from making things up or going off the rails.
The guardrails
AI without guardrails is a liability. Here’s what we built in:
Fixed service catalog — the AI can only recommend services we actually offer, at the exact prices we set. It cannot invent services, discounts, or timelines.
Conversation limits — maximum 5 exchanges before a recommendation. Prevents runaway conversations and keeps costs predictable.
Rate limiting — 10 conversations per hour per visitor. Prevents abuse and keeps our API costs under control.
Input sanitization — messages are limited to 500 characters and stripped of anything that could interfere with the AI’s instructions.
Hard boundaries — hostile messages, coding questions, illegal activity, and attempts to manipulate the AI all get a polite shutdown. No engagement.
Graceful fallback — if anything goes wrong (API down, network error, rate limit), the site automatically switches to a static quiz. You always get a recommendation.
Adversarial prompt testing — we test every prompt change against a suite of attacks: prompt injection, jailbreaks, hostility, offensive content, competitor fishing, and attempts to make the AI hallucinate services that don’t exist. If the guardrails break, the change doesn’t ship.
Audit trail — every conversation is logged with structured metadata: exchange count, recommendation made, confidence level, and latency. This lets us monitor quality, detect abuse patterns, and continuously improve the prompt based on real conversations.
No training on your data — we use the Claude API, which does not use your conversations to train AI models. Your business information stays private.
What this means for your business
This AI assistant is a small example of what we build for clients. The same approach — connecting AI to your specific business rules, data, and workflows — works for:
Customer intake — qualify leads, route inquiries, collect information before a human ever gets involved
Internal operations — answer employee questions about policies, process documents, generate reports
Client communication — draft emails, follow up on invoices, send reminders — all in your voice, all on autopilot
Data processing — categorize expenses, extract information from documents, reconcile records
The AI assistant on this site cost us about $0.015 per conversation to run. That’s less than two cents. The automation we build for clients has similar economics — pennies per task, replacing hours of manual work.
Built by engineers, not marketers
AI Jumpstart is a service by Agentic Studio Labs — an AI engineering firm that’s shipped 40+ production AI systems. We work across Claude, GPT, Gemini, and open-source models. We build direct API integrations, not just Zapier automations. And we’ve built AI security tools used by other companies to protect their own AI deployments.
When we automate your bookkeeping, sales pipeline, or content operations, you’re getting the same engineering rigor that built this assistant.
Want something like this for your business?
Book a free 30-minute Jumpstart call. We’ll figure out where AI can save you the most time — and what it would take to build.