B2B sales networking simulator

AI agents simulating B2B sales networking using real company data provided by whitecontext.com Two agents autonomously explore partnerships & make decisions. Built with Mastra + Nvidia NIM.

Comment

💡 The Idea

We're WhiteContext - we transform websites into structured B2B intelligence ($1/company API). We realized: companies have the data, but don't know how to USE it in sales.

Solution: AI agents that simulate B2B networking using real company intelligence.

🚀 What We Built

1. B2B Networking Simulator

  • Two AI agents (seller + buyer) explore partnerships
  • Autonomous conversations - agents decide when to stop
  • Real data from 220+ pharma companies
  • Autoplay mode + JSON export

2. Sales Training Simulator

  • AI prospect personas for sales practice
  • Custom scenarios or database companies
  • Realistic objections based on company context

🛠️ Tech Stack

  • Mastra: Agent framework + working memory
  • Nvidia NIM: Qwen 3 Coder 480B model
  • Next.js 15: Frontend + API
  • WhiteContext Data: 220 company intelligence dataset

💪 Key Challenges

  1. Agent identity confusion → Fixed with explicit role definitions
  2. Overly verbose responses → Enforced 2-3 sentence max with examples
  3. Structured output failures → Switched to text marker system [COMPLETE:reason]
  4. Conversations too long → Goal-driven termination, agents decide when done

🏆 Proudest Achievement

Autonomous Goal-Driven Termination

Turn 6: "Want to schedule a call?"
Turn 7: "Tuesday works!"
Turn 8: "Perfect! Talk then. [COMPLETE:meeting_scheduled]"
        → 🛑 Agent autonomously stops

No hardcoded rules. Agents reason about objectives and decide when conversation is complete.

📚 What We Learned

Multi-agent Mastra + WhiteContext data = emergent B2B behavior

  • Pre-loaded context 10x faster than tool calls
  • Working memory perfect for goal tracking
  • Real company data → specific, grounded conversations
  • Two agents with different goals → realistic negotiation dynamics
  • Agents can make autonomous decisions beyond pattern matching

🔮 What's Next

Immediate:

  • Analytics dashboard (conversation patterns, success metrics)
  • Quality scoring with Mastra scorers
  • WhiteContext API integration (live data)

Future:

  • Multi-agent networks (3+ companies)
  • Real-time sales coaching
  • Outcome prediction

📊 By The Numbers

  • 4 Mastra Agents (networking initiator/recipient, sales trainer, procurement)
  • 3 Custom Tools (prospect loader, BioCam context, accessibility)
  • # of Companies in intelligence database
  • 2-3 Sentences per agent response (enforced brevity)
  • 30 Max Steps for tool usage and iteration
  • ∞ Scenarios (database + custom configuration)

🎯 Use Cases

  1. Partnership Discovery - Generate and analyze B2B networking patterns
  2. Sales Training - Practice with realistic AI prospects
  3. Template Generation - Export conversations for outreach templates
  4. Communication Analysis - Study goal-driven conversation dynamics

Built With

  • claude-code
  • nextjs
  • nvidia-nim
  • qwen3
  • typescript
  • whitecintext.com
Share this project:

Updates