TradeAgent AI

Featured Privateai

AI business assistant for tradespeople — plumbers, electricians, HVAC pros. AI voice/SMS receptionist that answers calls 24/7, books jobs, sends quotes. Includes smart scheduling, CRM, invoicing, automated review requests, NPS surveys, recurring jobs, and multi-channel (Voice/SMS/WhatsApp/Email). PWA + dark mode.

Next.jsTypeScriptFastAPIPythonLangGraphTwilioDeepgram

A 24/7 AI receptionist for tradespeople, end-to-end on one stack.

The problem

Tradespeople — plumbers, electricians, HVAC techs — lose ~30% of inbound calls because they're on a roof, under a sink, or driving. Every missed call is a lost job to a competitor. Hiring a receptionist costs more than most one-person shops can justify.

The approach

Built a multi-channel AI agent (Voice / SMS / WhatsApp / Email) that answers calls, identifies the issue, books the right time slot against the tradesperson's calendar, sends a follow-up quote, and triggers a review request after the job. Voice path uses Deepgram for real-time transcription and a LangGraph state machine to handle the conversation flow — every branch (price quote, scheduling, escalation to human) is an explicit graph node, which makes the agent debuggable instead of a black box.

Tech decisions

LangGraph for the agent

Explicit state machine beats an open-ended LLM loop; every transition is auditable, errors don't cascade across turns

FastAPI on Python backend

Native async support for streaming audio; rich Python AI ecosystem available

Next.js + TypeScript frontend

Type-safe end-to-end with the API; PWA + dark mode out of the box

Deepgram for STT

Sub-300ms streaming transcription; better accent handling than the alternatives we tested

Twilio multi-channel

One vendor for Voice + SMS + WhatsApp keeps the integration surface small

Outcomes

AI receptionist answers calls 24/7, books jobs into real calendar slots
Multi-channel (Voice / SMS / WhatsApp / Email) with shared conversation history
Automated quote generation post-call with PDF delivery
Review request + NPS survey triggered on job completion
Recurring jobs + CRM + invoicing on one dashboard

What I learned

Building an agent as an explicit state graph instead of a chained-prompt loop pays for itself the first time a customer says 'why did it do that?' Every transition is logged and replayable. The Voice path is where 90% of the magic lives — text channels are easy; sub-second voice with proper barge-in is the hard part.