Minions.AI

Trade contractors — plumbers, HVAC technicians, electricians — lose a significant share of inbound leads simply because no one picks up the phone. They're on a job site. The call goes to voicemail. The prospective customer calls the next contractor on the list.

The second problem was content. Contractors need consistent, localized web content to rank in local search. Most don't have the time or staff to produce it regularly, and generic blog-writing tools produce content that's obviously templated and doesn't reflect local trade knowledge.

The voice pipeline was redesigned from sequential to overlapping streams. Rather than waiting for a full transcription before starting LLM generation, and waiting for the full LLM response before synthesizing audio, each stage fires as soon as it has enough data. Neural Voice Activity Detection replaced fixed silence timers, dropping end-of-turn detection from 800ms to 280ms. The first synthesized audio chunk reaches the caller in roughly 1.4 seconds from the end of their sentence.

For the editorial pipeline, the key architectural decision was to move all control flow out of the LLMs and into TypeScript. Each agent — researcher, drafter, critic — is a pure function that takes a typed state object and returns a typed delta. The state machine in TypeScript decides sequencing, retry logic, and escalation. LLMs can't improvise next steps or get stuck in loops, because they're never asked to manage execution — only to transform content.

The platform runs a fleet of voice agents that answer inbound contractor calls with natural conversational flow, qualify the caller's need, check technician availability via calendar API, and confirm bookings — without human involvement. Missed calls trigger an automated SMS within seconds containing a personalized follow-up.

The editorial engine runs on a scheduled basis, harvesting trade-specific customer pain points, generating structured drafts, running them through an independent critic agent that checks for banned clichés and unsupported claims, and publishing approved content to Cloudflare R2 for CMS staging. The mission control dashboard gives operators full visibility into which agents are active, what conversations are in flight, and what the per-inference cost is running at.

Contractors using Minions.AI handle inbound call volume without hiring additional front-office staff. The voice agents respond in under two seconds and maintain natural conversational flow across the full appointment booking workflow.

The editorial pipeline produces consistent, on-brand content across trade verticals without manual intervention. Every published article passes through the critic validation stage, which catches generic phrasing and unverifiable claims before anything goes live.