NuvAI

Status: In-progress product prototype. This case study documents the interaction model, service blueprint, and agent architecture. It does not claim production-scale validation.

Real-estate lead handling often starts with a gap: a prospect submits a form or calls, then waits for a generic follow-up while human agents juggle dozens of leads. Automation can answer instantly, but common qualification scripts optimize for extracting information quickly, often before a person understands who they are speaking to or has had a chance to explain what they need.

The design question was: Can an AI agent qualify a lead while giving the person meaningful control over pace, disclosure, and handoff?

I shaped the product and conversation experience across:

Conversation design and qualification flow

UX and interaction design for the agent interface

Service blueprinting across capture, qualification, and follow-through

Trust patterns for AI disclosure, pacing, and human escalation

Agent architecture decisions with Claude and n8n

The product treats trust as part of the interaction architecture, not a line of compliance copy. Three constraints guide every interaction:

Disclose first. The agent identifies itself as AI before asking questions.

Let the person set the pace. The conversation can move through neighborhood, budget, timing, or other needs in the order the lead chooses.

Keep a human handoff available. A person can pause, end the exchange, or move to a human without repeating their context.

The conversation starts with clear AI disclosure and an explanation of what the agent can help with. It then gathers qualification details through a flexible, person-led exchange rather than a fixed extraction script.

The agent can answer early questions about locations or process before asking for budget, timeline, or contact details. It records only the context needed for a useful next step and marks when the lead is ready for scheduling or human follow-up.

I mapped the end-to-end system across three stages:

Capture: A lead arrives through chat or form submission.

Qualification: The AI answers questions, gathers context at the lead’s pace, and identifies handoff readiness.

Follow-through: n8n routes the context, updates the CRM, and alerts the human agent for a warmer, more informed follow-up.

The blueprint made decision rights visible: what the agent can answer, what it should defer, and what information must survive the handoff.

An agent chat interface for disclosure, qualification, and pacing controls

A CRM dashboard concept for pipeline and qualification visibility

A service blueprint connecting the prospect, AI agent, workflow, CRM, and human agent

A 4-screen voice-layer exploration prototype, with a voice stack still under evaluation

The AI identifies itself at the beginning of the exchange. That gives people the information they need to decide how they want to participate, before the agent requests personal or financial details.

Fixed qualification sequences can make the agent feel like a form with typing indicators. NuvAI is designed to follow the lead’s priorities, then return to missing information when it is relevant.

A handoff only works if the prospect doesn’t need to retell the story. The system passes the conversation context and qualification state into the human follow-up so the next interaction can start with continuity.

Claude supports reasoning and the conversation layer. n8n orchestrates routing, CRM updates, scheduling, and alerts. A Vapi voice layer is being evaluated, but is not part of the current build. The stack follows the designed conversation rather than dictating it.

The current work establishes an interaction and service-design foundation. The next step is testing the core hypothesis: whether clear disclosure, person-led pacing, and a warm human handoff can produce accurate qualification context while preserving trust.

Validation should compare lead completion, quality of captured context, handoff readiness, and qualitative feedback against the humane-design constraints. Until that work is complete, the outcome is an in-progress prototype and a testable product hypothesis, not a performance claim.

NuvAI is an example of my approach to AI product design: make the system legible, give people control at the moments that matter, and design the operational handoff as carefully as the interface.