Author(s): Shahidullah Kawsar

Originally published on Towards AI.

AI Engineer Interview Preparation

A large e-commerce company is adding an AI assistant that can search products, compare prices, check inventory, and create support tickets. The first draft has one tool named search with a description that only says search records. During testing, the model sometimes searches customers when the user asks about products and sometimes searches products when the user asks about tickets. What is the best schema improvement?

After a set of MCQ-style scenarios, the article explains that reliable LLM tool use comes from designing tight, explicit tool schemas: split capabilities into narrow tools with clear boundaries and descriptions, use enumerations or strongly typed parameters to prevent invalid arguments, and require clarification (or safe validation errors) when inputs like identities or time are ambiguous. It emphasizes least-privilege and safety for destructive or sensitive actions by accepting constrained identifiers (not arbitrary paths) and keeping dangerous powers inside trusted application code. It also highlights structured, user-safe error responses for tool failures, domain-specific naming that scales as tool catalogs grow, architectural patterns like routing/dynamic tool loading to reduce candidate overlap, restrictive parameter types for consistent filtering, and schema testing focused on realistic ambiguous/underspecified requests rather than only happy paths.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI