Description
Summary **Senior AI Engineer to Productionize a Reusable Agent Platform** We are looking for an experienced AI engineer to transform an existing AI agent into a reusable, production-grade component for multiple client deployments. The agent already exists in a shared TypeScript monorepo and is integrated into a real application. However, it was initially built around one client and now requires architectural improvements, stronger reliability, and clear extension points for future deployments. This is not a greenfield chatbot or prompt-engineering project. We need someone who can assess an existing system, propose a pragmatic target architecture, and implement it. **About Sheled** Sheled is an AI deployment company focused on construction and other traditional industries. We build operational AI systems that integrate with each client’s data, documents, workflows, and software. Our goal is to reuse a strong shared platform while supporting controlled client-specific customization. **The work may include:** * Reviewing the existing agent architecture * Separating the shared agent core from client-specific tools, prompts, schemas, and workflows * Improving tool execution, structured outputs, and state management * Adding robust error handling, retries, timeouts, cancellation, and idempotency * Improving authorization and data-access boundaries * Adding tracing, logging, cost monitoring, and production debugging * Establishing automated tests and evaluation workflows * Documenting how the agent can be configured and extended * Migrating the current implementation to the improved architecture Our current stack includes TypeScript, Node.js, React, PostgreSQL, Supabase, LLM APIs, and a shared monorepo. **We are looking for someone who:** * Has personally built and operated an LLM agent or AI workflow in production * Has strong backend and software architecture experience * Understands tool calling, structured outputs, state, failure recovery, and observability * Can create reusable abstractions without overengineering * Is comfortable improving an existing TypeScript codebase * Writes maintainable, tested, and well-documented code Experience with queues, background workers, evaluation systems, authorization, or document processing is an advantage. **When applying, please answer:** 1. Describe one production AI agent or LLM workflow you worked on. What failed in practice, and how did you improve it? 2. How would you separate a shared agent core from client-specific behavior? 3. What would you inspect in the existing codebase before proposing a new architecture? 4. Please share relevant projects and clarify your personal contribution. We value pragmatic engineering over unnecessary abstraction. The goal is a dependable component that we can understand, operate, and reuse across real client deployments.