Voice AI (Illico) - Vapi Agent Engineering

Illico Voice AI — Pierre, a Vapi voice agent engineered to qualify training leads, handle CPF and Pôle Emploi objections, and book appointments at any scale

This is the most impressive feat of the Illico project: the design of the mind and voice of Pierre, an artificial voice agent built on the Vapi Voice AI infrastructure. Pierre is not a toy chatbot given a voice; he is a carefully engineered commercial agent, shaped through dozens of iterations of persona design, prompt engineering, and real-world call testing, until he became indistinguishable from a seasoned human call center agent — and, in many dimensions, genuinely better.

Realism and persona engineering

A voice agent is only as good as the illusion of humanity it sustains across the full length of a call. This one is built to that standard.

• Human vocal mimicry: vocal fluidity is parameterized to microseconds, tuned across the full Speech-to-Text, LLM, and Text-to-Speech pipeline. Turn-taking, backchannel sounds, breathing pauses, and reactive acknowledgments are all tuned so that the prospect never hears a "robot pause" that would break the illusion. Technically, Pierre is often indistinguishable from a seasoned human agent on the phone.
• Persona consistency: Pierre has a name, a background, a tone, and a set of conversational habits that are consistent across every call. He is not a generic AI voice — he is a specific, recognizable character designed for this commercial role.

Intensive qualification matrix

The intelligence behind Pierre is subjected to drastic system directives and prompt engineering that transform a general-purpose LLM into a highly specialized training advisor.

• Objection handling: Pierre is explicitly trained to methodically sweep the prospect's objections — lack of time, doubt about ROI, family situation, financing worries — without ever compromising the fluid thread of the discussion. Each objection has a planned response tree that sounds improvised but is engineered.
• Complex financing analysis: Pierre performs algorithmic real-time analysis of the prospect's financing solutions — CPF (Compte Personnel de Formation), Pôle Emploi funding eligibility, private budgets, employer co-funding — and adapts the conversation accordingly. This alone replaces a substantial amount of human expertise that is usually the bottleneck of a training sales team.
• Active qualification: Pierre doesn't just answer questions; he actively qualifies the prospect against the agency's ideal customer profile, identifying whether the lead should be booked for a closing call with a human advisor, nurtured, or respectfully declined.
• Action passing: once Pierre has identified a qualified prospect and aligned on a time, he orchestrates the actual appointment booking through the Make.com tool-calling layer (see the parallel project), turning intent into a locked CRM slot inside the same call.

Why Pierre is commercially different

Most voice AI deployments fail commercially because they fail in one of three ways: they sound robotic, they handle objections poorly, or they cannot actually close the loop into a CRM. Pierre solves all three in one coherent product. He is an unparalleled salesperson in his specific domain: tireless, ultra-reactive, consistent, and capable of handling a volume of thousands of simultaneous calls without any of the quality drift a human team would experience at scale.

Engineering discipline behind the scenes

• Iterative prompt refinement tied to the analytics dashboard (see parallel project): every flagged failure becomes a candidate prompt adjustment.
• Guardrails: Pierre is explicitly prevented from making promises outside his remit, from discussing competitors, or from improvising on pricing.
• Escalation paths: when a call goes outside Pierre's territory, he gracefully escalates to a human — without breaking persona and without losing the prospect's trust.

The delivered outcome

• A voice agent that handles training lead qualification, CPF and Pôle Emploi analysis, objection handling, and appointment booking — end to end, on every call.
• Quality that is indistinguishable from a seasoned human agent, at a fraction of the cost.
• Effectively infinite scalability: thousands of simultaneous calls with zero degradation.

Technology stack

• Vapi as the Voice AI infrastructure.
• Speech-to-Text and Text-to-Speech components tuned for human-grade conversational fluidity.
• Linguistic prompt engineering as a core discipline for persona, objection handling, and qualification logic.
• Artificial agent architecture integrated with CRM tool-calling for real-world commercial action.

---

Ready to automate your business? Discover our AI Automation service →