Project 12 / AI & support
Conversational AI Operations
A chat-and-voice operations layer with consent-aware lead capture, safe handoff and transcript regression QA.
01 / Problem
A chatbot needs state, safety and measurable quality—not only fluent text—to be useful in sales or support.
02 / System
Designed as a controlled operational pipeline.
Two workflows route lead and support conversations, create CRM-ready leads only with consent, block sensitive actions and score regression transcripts.
Consented lead captured
Voice delivery metadata produced
Strong and weak transcripts distinguished
03 / Reliability controls
What keeps the system honest.
04 / Verification
Actual n8n webhook
Actual n8n webhook passed lead, replay, voice, sensitive, validation and model-outage paths.
This is a locally verified engineering demonstration using synthetic data and controlled service adapters. It is not described as a production client deployment unless live evidence is added later.
Inspect the implementation.
Sanitized exports and test artifacts from this demonstration. No credentials or client data are included.
Lead, support, consent and handoff routing.
Open file JSONTranscript QA workflowRegression scoring for conversation quality.
Open file JSONPrompt contractVersioned scope, safety and output requirements.
Open file JSONConversation fixturesChat, voice, replay and failure cases.
Open file JSONMachine-readable executionActual webhook statuses and final state.
Open file MDVerification reportClaims, scenarios and evidence limits.
Open file