AI Operations Intern

**About DirectHedge** DirectHedge is building the operating system for physical lumber trading. Our platform combines an order management system (OMS), an anonymous physical lumber exchange, and an AI layer called DirectIntelligence (DI) that automates the operational workflows traders and buyers deal with every day. We’re a small, fast-moving team based in Vancouver, and everything we ship goes live to real users. **The Role** This isn’t a sandbox project. DirectHedge already has a live MCP (Model Context Protocol) server with 30\+ tools, a working OMS portal, and five AI agents in development. You’ll be building on top of what exists — extending tools, refining agent behavior, writing the instruction files that make AI agents work correctly, and connecting the pieces together into a production system. **What You’ll Work On** **MCP Tool Development:** Our MCP server already exposes tools for trading documents, counterparties, reference data, and authentication. You’ll add new tool definitions, improve existing ones, write test coverage, and ensure the schema documentation is accurate. When gaps surface, you’ll file tickets in Linear and work with the backend team to close them. **AI Agent Refinement:** DirectIntelligence has seven agents handling invoice processing, customer invoicing, collections, cash flow, inventory, and more. You’ll refine their system prompts, tune tool-use behavior, build evaluation workflows, and potentially architect new agents as needs emerge. **Skill \& Instruction Authoring:** The agents live or die by their instruction files (markdown-based skill definitions). You’ll write, test, and iterate on these files — treating them as code, not documentation. **Backend Exposure (Stretch):** If you prove yourself, you’ll get access to the backend codebase to add or modify API endpoints that the MCP tools and agents depend on. This is earned, not given. **What We’re Looking For** **Required:** – Comfortable with TypeScript/JavaScript. Working knowledge of Python is a plus. – Practical understanding of how LLMs work — prompt engineering, tool-use patterns, structured output parsing. You’ve built something with an LLM API, not just chatted with one. – Strong written communication. Half this job is writing precise instructions that make AI agents behave correctly. Sloppy writing \= sloppy agents. – Can read an API spec and write clean documentation without hand-holding. – Enrolled in or recently graduated from a computer science, software engineering, or related program. **Strong Signals (Not Required):** – Experience with MCP, LangChain, or similar agent/tool-use frameworks. – Familiarity with Linear, Git-based workflows, and working in small teams that ship fast. – Any exposure to logistics, supply chain, or commodity trading software. Rare in an intern, but valuable. **What You’ll Get** – Real production experience. Everything you build goes live to actual users in a real lumber trading operation. – Direct mentorship from the founder. No layers of middle management. You’ll work alongside the person who designed the platform. – Exposure to the full stack — from AI agent design to backend APIs to product decisions. – A portfolio piece you can actually talk about. This isn’t a toy project — it’s a commercial platform in an industry that’s never had one. **Details** **Duration** 10–12 weeks (Summer 2026\) **Location** In-office, Vancouver, BC **Reports To** David Topps, Managing Director Pay: From $25\.00 per hour Work Location: In person