• GIF

  SalesforceAgentInsight

🧠 AgentInsight – Project Story

✨ Inspiration

As autonomous agents become increasingly common in business environments, I wanted to explore a way to make them more trustworthy and accountable. The idea of an agent that could evaluate other agents' responses — almost like a peer-review system — inspired the concept for AgentInsight. Salesforce’s Agentforce provided the perfect context to build this kind of introspective, intelligent system.

🛠️ What I Built

I built AgentInsight, an autonomous agent prototype using the Agentforce architecture. The system simulates:

• An agent receiving a prompt
• A second agent responding
• AgentInsight analyzing and scoring that response based on clarity, correctness, and self-awareness

I created a mock Apex handler, a Lightning UI simulation, CSV data logs, and an interactive Plotly visualizer for analyzing agent performance.

💡 What I Learned

• Designing for agent-to-agent interactions opens up exciting possibilities for internal validation and QA.
• Trust in autonomous agents can be enhanced through introspection and score-based evaluation.
• Even without full Salesforce API access, it's possible to create a meaningful prototype using mock logic and simulation.

🚧 Challenges

• No access to a live Salesforce org during development meant I had to simulate all Agentforce logic.
• Balancing the concept of “self-awareness” in a way that’s both measurable and business-relevant was surprisingly difficult.
• Ensuring the visual and technical components remained lightweight and demo-ready for hackathon constraints.

✅ Outcome

AgentInsight is a working prototype ready to plug into real Agentforce environments. It shows how agents can self-reflect and evaluate peers to build a more robust, transparent, and auditable AI-powered workflow.

Built With

• salesforce