Manual analysis and emotional bias are the ultimate performance killers in trading. But how do you automate complex strategies without handing over control to an unpredictable "black box" AI? The solution lies in moving away from monolithic systems toward Agentic Software Engineering.

In this session, I will demonstrate how to build a high-performance digital trading desk using the Semantic Kernel Agent Framework. We will leverage a team of specialized AI agents that don't just follow static scripts but dynamically cooperate in a Group Chat to validate trading decisions.

Technical Deep-Dive:

Multi-Agent Orchestration: Leveraging AgentGroupChat to manage speaker selection and termination strategies. Who speaks when, and who has the final word?

Separation of Concerns: Designing specialized agents for market scanning (Signal), mathematical risk assessment (Kelly Criterion/ATR), and final execution.

Human-in-the-Loop 2.0: Implementing a secure approval chain where the AI handles the heavy lifting, but the human retains ultimate authority.

Observability & Audit: Using Application Insights to track the "thought process" of your agents—essential for post-trade analysis and compliance.

The Live Demo: We are going live! I will demonstrate the full lifecycle: from detecting a technical breakout and the discursive risk assessment within the agent chat to the final generated order proposal. Watch the real-time interplay between code and AI—complete with live performance monitoring and log analysis.

Takeaways: You will walk away with a battle-tested blueprint, including code templates for market integrations and risk engines. This isn’t about "get rich quick" schemes—it’s a deep dive into how multi-agent systems are revolutionizing software architecture in the financial sector.