Build AI

The Problem

Businesses wanting a capable AI assistant faced a choice between rigid template bots that could not hold a useful conversation, and bespoke development that required an engineering team they did not have.

The gap was not conversation quality on its own. A support bot that can explain a refund policy but cannot issue the refund still leaves the work with a person, so the cost the business wanted removed stays exactly where it was.

What We Built

One platform core, deployed as four vertical products: DentalOS for dental and medical clinics, CommerceAI for e-commerce support, Restaurant Agent for restaurant operations, and HirePilot for recruitment.

The core carries what every one of them needs — isolated multi-tenant data with per-organisation configuration and branding, SSE-driven real-time sync across connected clients, JWT auth with role-based access and audit logging, and a knowledge base backed by pgvector search.

Each vertical then adds its own domain: an FDI-notation odontogram and multi-branch appointment diary in DentalOS; refunds, order tracking and cart recovery in CommerceAI; menu search, complaint routing and feedback collection in Restaurant Agent; an applied-to-hired applicant pipeline with structured intake forms in HirePilot.

Agents That Act

The agents are wired to the systems of record, not parked in front of them. A supervisor agent routes an incoming query, and the agent it hands to can process a refund, book an appointment, look up an order or move a candidate through the pipeline.

One agent serves every channel a customer arrives on — WhatsApp, Instagram, Facebook, email and the web widget — through a single inbox, so a conversation that starts on one does not restart on another.

Outcome

The line ships commercially from buildaierc.com, with each product deployed on its own subdomain and running against live customer data.