Raar

Ali and I started this out of boredom. We had a lot of free time, kept hearing AI was the thing that could automate any kind of work, and wanted to find out what that actually meant.

So we walked into every shop on Penn State’s campus and asked the same question: what’s a daily, repetitive task you wish you didn’t have to do? Crude market research, but it surfaced a real one fast. A barbershop owner told us his biggest problem was the phone: calls he couldn’t pick up, no shows he couldn’t chase.

We went home and started learning how any of this was supposed to work. Neither of us had any prior knowledge of AI, automations, or coding, so we figured it out one piece at a time. After a while, we had a working demo of an AI receptionist for him.

We never plugged it in. And by then we were already somewhere else, more interested in what we’d just made than in selling it back to him. We moved on to a consumer version of the idea.

Kendall is an AI receptionist for service businesses: any business that runs on appointments where the owner is also the person who has to answer the phone.

She handles the phone so the owner doesn’t have to. Bookings, callbacks, reschedules, reviews, schedule changes, cancellations. Every customer interaction that runs through a call.

What she actually replaces is the front desk a small business can’t afford to hire. A receptionist who’s there every hour, knows every customer, never misses a call, never forgets a preference.

If you want to learn what she’s capable of, call her →

We took everything we learned from Kendall Business and reshaped it for one person instead of one business.

What pulled us toward the personal version was curiosity. The tech kept getting more interesting the more we built, and we wanted to see how far it went when the user wasn’t a business but a person designing their own assistant.

Kendall Business is one system, set up per shop: same core, different details. MyKendall didn’t have one shape; it had as many as users would design. So most of the work was finding the right formula: how the way someone set up their assistant, the voice and behavior of the agent, and what was actually running it had to come together so a user’s choices in onboarding produced an assistant that showed up on the phone the way they designed it.

MyKendall is your own AI assistant. You pick her voice, her personality, and what she knows about you: who you are, what you do, who you’ll and won’t take calls from.

It’s live and fully working. Mostly, it’s just there for people to play with.

VAPI↗ handles the live STT → LLM → TTS loop under 800ms. ElevenLabs↗ is the voice, Twilio↗ carries the call.

Several agents are orchestrated in LangGraph↗ (greeter → booking → handoff for Kendall Business; triage → tool call → memory write for MyKendall). Reasoning is OpenAI GPT-4o↗.

Memory is Supabase↗ with pgvector. Every call writes, every future call reads.

Background work (email summaries, calendar syncs, scheduled tool calls) runs on Trigger.dev↗.

n8n↗ is the integration spine: calendar, booking, payment systems.

Web is Next.js↗ + TypeScript↗ on Vercel↗.