THIS 23-YEAR-OLD BUILT AN AI RECEPTIONIST THAT NEVER MISSES A CALL
Every day, restaurants lose customers simply because nobody answers the phone.
So instead of hiring another receptionist, this 23-year-old built an AI voice agent.
Call the restaurant, and the AI answers instantly.
It books your table, changes reservations, answers common questions, and works 24/7—without anyone realizing they’re talking to AI.
He charges restaurants $2,500 to set it up and $500/month to maintain it.
After signing dozens of local businesses, his AI agency is now generating over $120,000 in monthly recurring revenue.
This isn’t about replacing people.
It’s about making sure businesses never miss another customer.
Would you trust an AI to book your next restaurant reservation?
Follow @Dexonfxf for more AI business ideas.
YOUR LOCAL AI MAP STARTS AT $249 JETSON, A USED DELL OPTIPLEX MICRO WITH AN I7 10TH GEN AND 16GB RAM SELLS ON EBAY FOR $30-50 AND RUNS LLAMA 3.2 7B AT 7 TOKENS PER SECOND, THE NEW SUB JETSON TIER ZERO
01:11 he scrolls eBay, "if we do sort by price some of the cheapest are 30 to 50 dollars, considerably cheaper than most of the raspberry pi fives"
a Dell OptiPlex 5070 micro from corporate decommission lots sells for $35 with an i7 9700T, 16GB DDR4 socketed RAM, an M.2 NVMe slot and a 35 watt TDP
ollama on debian runs llama 3.2 3B at 12 tokens per second on integrated UHD 630 graphics, llama 3.2 7B at 7 tokens per second, mistral 7B at 6 tokens per second, all on CPU+iGPU with no discrete card
a raspberry pi 5 16GB ships at $120 plus case and adapter, the OptiPlex undercuts it by 70% with 8x more compute, 2x more RAM and a real M.2 slot for a 2TB NVMe upgrade
your serie maps $249 Jetson as the entry point, eBay decommission lots just opened a sub jetson tier at $35-50, the local AI floor dropped to less than a single openai monthly bill
the window is open, follow and bookmark before it closes
A $700 HOMELAB CAN DO WHAT 5 CLOUD SUBSCRIPTIONS KEEP REPACKAGING AS "AI INFRA"
The video is not about a guy flexing cables in a rack.
It is a small home stack: router, server box, monitoring screen, clean networking and a workstation plugged into infrastructure that runs 24/7.
That matters because agents are useless if they only live inside a browser tab you remember to open.
The real setup is always-on compute, storage, networking, remote access, automations and monitoring in one place.
Run an agent overnight. Host your files. Keep logs. Trigger workflows. Access your machine remotely. Stop renting every tiny piece of your workflow from another SaaS dashboard.
The article is pointing at the same operator shift: serious AI use is becoming infrastructure, not just prompts.
People are still comparing models while builders are setting up the layer the models actually run on.
The model gives you intelligence.
The homelab gives it a place to work when you are not watching.