Twin AI.
Duration
Dec 2025-Jan 2026
Location
Mountain View, CA
Task
AI Adoption
AI Agent Onboarding
Dashboard
Building agentic AI for automated CRM @Trademarkia
During my internship at Trademarkia (a legal-tech firm in Silicon Valley), we rethinks client intake for real estate and legal professionals. I designed workflows for Twin AI, focused on enabling the trustworthy adoption of AI-mediated client intake.
Key Contributions:
– Designed and developed an AI Sandbox enabling experts to fine-tune tone and RAG knowledge, driving 55% feature adoption.
– Closed the loop from intent detection to CRM entry, reducing processing time by 70% and increasing high-value leads by 25%.
– Built a node-based AI orchestrator that turns client intake into a 24/7 automated pipeline, boosting conversion by 35%.
Impact & Value
An AI Sandbox for always-on intake SaaS
I Crafted a node-based AI orchestrator to automate client intake for real estate and legal professionals. Visualized prompt engineering, boosting config efficiency by 60% and increasing conversion by 35% vs. manual setups, achieving 42% higher client conversion. The workflow I designed scaled from a single voice-based acquisition touchpoint to an AI operating system for trillion-dollar regulated industries.
Highlight:
– Turned client intake into a 24/7 AI pipeline (+35% conversion)
– Eliminated 70% manual ops with end-to-end automation
– Scaled expert-driven AI systems (+55% adoption)
MAIN CHALLENGE
Enabling professionals to trust an AI agent to represent their identity, expertise, and tone in high-stakes client interactions.
Transforming relationship-driven client intake into AI-mediated automation is less a technical challenge than a trust and behavior shift. Because the agent speaks as them, adoption depends on whether it feels like trusted delegation rather than loss of control.
Goal
Design a low-friction onboarding workflow that feels human, is effortless to manage, and closes the loop from call to system.
My role was to translate this mental-model shift — from “I answer calls” to “my agent answers for me” — into a clear, controllable workflow.
Design Progress.
FEATURE 1
Segmented Voice Agent Setup for Scenarios
In the first-time onboarding scenario, where agents need to set up their first AI twin quickly, we designed preset and reusable voice options to reduce setup friction.
As agents move beyond onboarding into more advanced usage, where trust and personal branding become critical, we introduced high-fidelity voice cloning to capture each agent's unique timbre and emotional inflection.
FEATURE 2
Low-Friction Flow to Equip AI with Property Knowledge
Users can select specific properties from their database to "teach" the agent the relevant details. By linking pricing, location, and size directly to the agent, the AI operates with a grounded, business-accurate knowledge base, rather than generic conversational intelligence.
FEATURE 3
Visual Script Editor to Control AI Conversations
To support users without technical backgrounds, we designed a visual, node-based script editor.
Granular Control: Users can toggle questions on/off, drag to reorder the conversation flow, and edit specific wording to match their personal style.
Visual Mapping: A node-based canvas showing the journey from the "Start" greeting to "Lead Qualification".
FEATURE 4
Embedding AI into CRM for Lead Management
The workflow culminates in a high-efficiency Leads Dashboard, ensuring a seamless handoff from AI to Human.
AI-generated conversation summaries and a lead claim priority score eliminate the need for agents to replay long recordings, dramatically improving follow-up efficiency and decision-making speed.
Evaluation.
TESTING & EVALUATION
Copilot study with 11 U.S. professionals showed strong interest and generated actionable feedback.
In moderated virtual sessions, professionals configured an AI agent, reviewed onboarding outputs, and evaluated performance in simulated client calls. Feedback highlighted strong interest alongside needs for clearer onboarding, greater behavioral control, and higher fidelity to professional tone and expertise.
ITERATION
Enable ongoing visibility of agent behavior
The initial dashboard focused on agent setup and status. We evolved it to surface ongoing activity, call outcomes, and conversation previews, making AI actions visible and inspectable.This improves trust by showing what the agent is doing on the user’s behalf.
ONGOING IMPACT
The story does not end here...
The workflow I designed scaled from a single voice-based acquisition touchpoint to an AI operating system for trillion-dollar regulated industries.
Twin AI began as Trademarkia’s voice intake agent and became the foundation for GetBill — the company’s new AI operating layer for regulated industries. As it integrated into CRM, communication, and onboarding workflows, the agent expanded from a single touchpoint into continuous operational support.
Reflection.
REFLECTION 1
1.Enterprise AI adoption is workflow change, not feature addition
Transforming relationship-driven client intake into AI-mediated automation is less a technical challenge than a trust and behavior shift. Because the agent speaks as them, adoption depends on whether it feels like trusted delegation rather than loss of control.
REFLECTION 2
2. Control perception outweighs automation power
Transforming relationship-driven client intake into AI-mediated automation is less a technical challenge than a trust and behavior shift. Because the agent speaks as them, adoption depends on whether it feels like trusted delegation rather than loss of control.