Senior AI Backend Engineer – Fix OpenAI Tool Calling Loop (Laravel SaaS)

Job Details

Description

Summary Description We are building an AI-powered salon management SaaS platform. Our AI assistants have access to multiple internal tools (bookings, customers, revenue analytics, dashboard metrics, staff, services, etc.). The backend is built with Laravel (PHP) and uses OpenAI. We are looking for an experienced AI backend engineer to improve our current AI agent architecture. Current Problem Our AI can call tools, but the orchestration is implemented incorrectly. Current flow: User asks a question. LLM requests a tool. Backend executes the tool. Tool result is processed separately. Final response is returned. Because of this: the main LLM never reasons over the tool output; it cannot combine multiple tool results; responses are generic instead of data-driven. What We Need We want to implement a proper multi-step tool-calling loop: User ↓ LLM ↓ Tool Call ↓ Backend executes tool ↓ Tool Result ↓ LLM ↓ (optional additional tool calls) ↓ Final answer The AI should be able to call multiple tools before generating the final response. Scope of Work We need someone who can: review the current architecture; refactor the tool-calling flow; implement a proper OpenAI tool loop; support multiple sequential tool calls; return tool outputs back into the same conversation; improve conversation state management; implement structured JSON tool schemas; prevent infinite tool loops; improve error handling; improve logging and debugging. Bonus if you have experience with: LangGraph LangChain OpenAI Responses API Langfuse AI agent observability Tech Stack Laravel (PHP) OpenAI API MySQL REST APIs Required Experience Please only apply if you have experience with: OpenAI Tool Calling / Function Calling AI Agents Multi-step Agent Workflows Agent Orchestration LLM Context Management Laravel / PHP Backend When Applying Please answer these questions: Have you built AI agents using OpenAI Tool Calling? Have you implemented multi-step tool loops where the model can call multiple tools before answering? Have you worked with Laravel? How would you implement the following flow? LLM ↓ Tool Call ↓ Tool Result ↓ LLM ↓ Additional Tool (if needed) ↓ Final Answer Have you used LangGraph, LangChain or similar frameworks? Please share links or screenshots of similar AI agent projects. Based on this description, what is your rough estimate (hours and cost)? Important We are NOT looking for: Prompt Engineers Chatbot builders Make.com / Zapier automation experts No-code AI developers We are looking for someone with real experience building production AI agents with tool calling and orchestration.

Skills

Notification History

User Actions