Developer builds Azure Functions scam-checking agent, finds agentic model fails at orchestration
A developer built a weekend project using Microsoft's Azure Functions hosted skills to verify whether online shops are legitimate, drawing on signals like domain age, HTTPS checks, archive history, and web search results. The tool used two tiers of evidence: keyless verifiable data from RDAP and the Internet Archive, and reputation signals via Tavily web searches targeting Trustpilot and fraud-related queries. The initial version gave the AI model full orchestration control over four verification tools, following standard agentic design patterns. This approach quickly ran into critical failures, including rate limit errors on GPT-4.1 and a context window overflow caused by a single webshop query — even after switching models and increasing capacity. The developer concluded the fix was architectural rather than technical, deciding to move orchestration logic into code and reserve judgment for the model, a pattern he now considers a default starting point.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)
Log in to join the discussion and vote.
Log in