Kevin Madura

About Kevin Madura

Kevin Madura is part of the AI engineering team at AlixPartners, the global management consulting firm focused on bankruptcy and restructuring, business and technology advisory, and risk and compliance. Before AlixPartners he was a cybersecurity principal at IBM. He has previously spoken at AI Engineer about using DSPy for structured outputs.

In an AI Engineer talk with Natalia Connolly, he explains how AlixPartners used Claude Code on a post-merger procurement analysis covering more than 40,000 contracts, and why the agent's first answers, which looked right, were off by about $800,000.

Notable Positions

On Structuring Data Up Front

For engagements where the needed fields are known in advance, such as rate tiers and rebates, the team extracts them into structured tables first (using DSPy on page images for charts) and lets Claude Code work inside that framework, instead of letting it loose on raw documents.

On Business Logic in Tools

Deal-specific tools like "get rate card page" or "get vendor spend" carry known gotchas in how they query the data, and get packaged into custom MCP servers so knowledge accrues across an engagement. See Context Engineering.

On Failing Loudly

When an access-denied error came back as an empty or zero value, the agent treated it as a real invoice amount. Tools should tell the agent what went wrong so it can recover or go to another authoritative source.

Key Quotes

• "You just have a defined set of operations that you're comfortable with, and you block everything else." (on governing agent data access)
• "context is still king" (closing takeaway)

Related Reading

• Context Engineering - Giving agents the right business context
• Natalia Connolly - Co-presenter at AlixPartners