AI Agent Developer - Build LLM Agent MVP with Tool Integrations

Title: Build an MVP AI Agent (Claude + Notion) — Internal "Advisor" Agent for a Managed-AI Services Firm Category: AI / ML · Chatbot & Agent Development Engagement: Fixed-scope first project, strong potential for ongoing work Budget: (left out — discuss after reviewing the brief) Who we are We're Everything Else (legal name We Do Everything Else) — a done-for-you AI services firm in Ottawa, Canada. We find where AI saves our clients time or grows their revenue, build the tools, and manage them ongoing. The owner never has to become the AI buyer, builder, or babysitter. Internally, our entire delivery model runs through one persona we call Addy, the Advisor — a single point of contact who coordinates eight service areas for each client. Right now Addy is just a carefully written prompt we run by hand. We want to turn her into a real, working internal agent. How we work (and how this job works) We prove value on one focused project before asking anyone for a longer commitment. We run freelancers the same way. This is a scoped, fixed-outcome first project. Do it well and there's months of follow-on work — building out the eight specialist agents that sit behind Addy, one at a time. What you'll build — the MVP We're Addy's first client. Before we run Addy for anyone else, we use her to run Everything Else itself. So the MVP isn't a demo — it's the working Advisor agent we operate day to day. Addy sits at the hub and coordinates all eight service spokes for that one client (us). In plain terms, Addy needs to: Connect to our Notion workspace and read the client brief, meeting notes, and pipeline status. Be the single coordinator across all eight spokes — Marketing, Web, Reputation, Sales, Support, Operations, Reporting, Doc Automation — routing each piece of work to the right one. In the MVP the spokes are simulated by Addy herself (we give you each spoke's definition and quality bar); real, separately-built spoke agents come later. Draft our standard deliverables from templates — an Opportunity Map, a Project Debrief, and a Monthly Update — filled with real context. Keep a human in the loop: Addy drafts, we review and edit, we ship. No actions that reach anyone outside our team on their own in this phase. Internal-only, human-in-the-loop, one operator. We'd rather have the hub working cleanly across all eight spokes than one spoke fully automated and the rest missing. Tech context Notion is our single source of truth. The agent reads from and writes to Notion, not a separate database. We default to Claude but want the model swappable — don't hard-wire us to one vendor. Everything you build belongs to us — full source, documented, no proprietary platform we get trapped in. We hold our clients to that standard; we hold our vendors to it too. Keep the stack simple and maintainable by a small team. Claude, Notion, and Vercel are already in our world. We have a recommended direction, but we want your proposal on the architecture. That's part of what we're hiring for. Where this goes Addy coordinates eight "spokes" — Marketing, Web, Reputation, Sales, Support, Operations, Reporting, and Doc Automation. The MVP is Addy herself. The next phase is building those eight as their own agents behind her. A strong MVP turns into a long working relationship. You're a strong fit if you Have shipped real LLM agents — not chatbot demos — with tool/function calling. Know the Notion API (or have integrated comparable structured workspaces). Are genuinely good at prompt and persona engineering and can keep an agent on-voice. Write clean, documented, handover-ready code. Communicate plainly and hit dates. We commit to two-to-three-week delivery for our own clients; we expect the same discipline from partners. Not a fit if you'd lock us into a proprietary framework, you've only built no-code Zapier/Make flows, or you can't show working agent code. To apply — answer these (skip the generic cover letter) Link to one agent you've built that uses tools/function calling. What did it do, and what was hard about it? How would you approach Notion as the agent's read/write memory — official API, MCP, or something else — and why? What interface would you build for an internal, single-operator agent like this, and why? (Web app, CLI, something on top of an existing platform — your call, defend it.) How do you keep an LLM agent reliably on-voice and on-format across many sessions? Your first instinct on timeline and how you'd phase the MVP.