SalesarenaAI

• 

  Demo Salesarenaai

• 

  Clear Architecture Diagram

Project Name SalesArena AI

Tagline Train like a closer. Master sales conversations with AI.

Inspiration Gen Z is entering the workforce with strong digital skills but limited experience handling real sales conversations. Most sales training today is either:

static courses scripted roleplay sessions or text-based chat simulations None of these recreate the pressure and unpredictability of real buyer conversations. The most difficult part of sales is handling objections in real time: “Your competitor is cheaper.” “This is out of our budget.” “We already have a vendor.”

SalesArena AI was built to solve this problem. We wanted to create a live AI training environment where users can practice sales conversations against realistic buyer personas that interrupt, challenge, and negotiate just like real prospects.

Using Gemini Live multimodal capabilities, we created an immersive roleplay simulator that helps Gen Z develop real-world sales skills.

What it does SalesArena AI is a real-time AI sales objection simulator that trains users through live voice conversations with AI-powered buyer personas.

Users can practice selling a product while the AI behaves like a skeptical prospect and introduces objections throughout the conversation.

The agent:

• listens to the user's pitch • interrupts naturally with objections • asks challenging follow-up questions • simulates realistic buyer personalities • analyzes the user's responses • generates a structured sales skill scorecard

After each session the system evaluates: confidence clarity objection handling value framing closing ability Users receive personalized coaching feedback so they can improve their sales technique. SalesArena AI turns sales training into an interactive learning experience instead of passive theory.

Key Features Live Voice Conversations Users speak naturally with the AI buyer using real-time streaming voice interaction powered by Gemini.

Realistic Buyer Personas The AI can simulate different prospect personalities such as: The Skeptical CFO The Budget Guardian The Busy Executive The Curious Product Expert The Aggressive Procurement Manager Each persona behaves differently during conversations. Objection Handling Simulation

The AI injects realistic objections during the conversation including: price objections competitor comparisons ROI challenges risk concerns Natural Interruptions

Using Gemini Live capabilities, the agent can interrupt the user naturally during their pitch to simulate real buyer behavior.

Sales Skill Scorecard

After each session the system generates a structured evaluation including:

Confidence Objection Handling Value Framing Clarity Closing Skills

Coaching Feedback

Users receive personalized suggestions on how to improve their responses.

Vision Mode (Optional)

Users can share a slide or product page and the AI can analyze and critique the visual sales material.

How we built it

SalesArena AI uses Google's multimodal AI stack to create a live conversational agent.

AI Model

Gemini (via Vertex AI)

Agent Framework

Google Agent Development Kit (ADK)

Live Interaction

Gemini Live API for streaming voice conversation and interruption handling.

Backend

Python FastAPI service that manages:

agent orchestration

session tracking

scoring logic

Frontend

React web interface that includes:

voice input

conversation transcript

training dashboard

scorecard visualization

Google Cloud Services

Cloud Run Hosts the backend service.

Vertex AI Runs Gemini models for the agent reasoning.

Firestore Stores conversation transcripts and training scores.

Cloud Build Automates deployment.

Architecture

User Browser ↓ React Frontend (voice streaming) ↓ FastAPI Backend (Cloud Run) ↓ Gemini Live API / Vertex AI ↓ Agent Development Kit (persona logic + tools) ↓ Firestore (session data + scoring)

Challenges we ran into

One challenge was designing the interruption behavior so that the AI could challenge users realistically without breaking the conversation flow.

Another challenge was building realistic buyer personas that respond differently depending on how the user answers objections.

We also focused on grounding the agent’s reasoning so that feedback is structured and consistent, rather than generic AI advice.

Finally, optimizing real-time streaming interactions required careful orchestration between the frontend voice system and the Gemini Live API.

Accomplishments that we're proud of

We built a fully interactive AI roleplay training environment that behaves like a real prospect during a sales conversation.

The system demonstrates:

live multimodal interaction

natural conversational interruptions

realistic roleplay personas

structured performance analysis

SalesArena AI transforms sales training from static learning into a hands-on immersive experience.

What we learned

This project showed how powerful live multimodal AI agents can be when applied to real-world skill training.

Instead of replacing human interaction, AI can simulate realistic environments where people can safely practice difficult situations such as negotiation and persuasion.

We also learned that designing believable AI personas requires both technical engineering and behavioral modeling.

What's next

Future versions of SalesArena AI could include:

team-based sales training environments

AI-generated buyer personalities from real CRM data

enterprise sales coaching dashboards

integration with CRM platforms like Salesforce or HubSpot

multiplayer roleplay sessions for sales teams

Our vision is to build the AI Sales Gym for the next generation of professionals

Built With

• agentsdk
• fastapi
• gemini
• react