AI Engineer — Build an AI Marketing Agent (Claude API + Meta/Google Ads)

AI Engineer — Build an AI Marketing Agent (Claude API + Meta/Google Ads) Project overview I'm building an autonomous AI marketing agent in the spirit of AdAI / AdGent — an agent that connects to ad platforms (starting with Meta Ads, then Google Ads), audits campaigns, surfaces wasted spend, and executes optimizations with human approval. Think "chat with your ad accounts, get a senior performance marketer in a box." Looking for a senior engineer to lead the build of the MVP over ~3–4 weeks. contract, remote. What you'll build - Tool/integration layer wrapping Meta Marketing API (campaigns, ad sets, ads, insights, creatives, audiences) as clean function-calling tools for the Claude API - Agent loop using Anthropic's SDK with multi-turn tool use, skill/playbook loading, and approval gates for any mutation that spends money - A small library of Markdown "skills" encoding our optimization playbooks (campaign audit, creative refresh, budget reallocation, search term analysis) - OAuth + multi-tenant account connection, encrypted credential storage, audit log of every agent action - Background workers for scheduled audits (nightly/weekly) and Slack/email digest delivery - Thin chat UI (Next.js or similar) streaming from the Anthropic API - Google Ads API integration as phase 2 Must-have experience - Shipped production code using the Anthropic SDK (or OpenAI function calling) with multi-turn tool use — please share a code sample or repo - Direct experience with at least one major ad platform API: Meta Marketing API or Google Ads API (please specify which, and what you built) - OAuth 2.0, refresh token handling, multi-tenant SaaS patterns - Solid Python or TypeScript, Postgres, a background job runner (Temporal, Celery, BullMQ, etc.) Nice to have - MCP (Model Context Protocol) server experience - Worked through Meta App Review or Google Ads developer token approval - Performance marketing domain knowledge (you've actually run paid ads) - pgvector or other embedding store experience How I'll screen Please include in your proposal: 1. A link to a tool-use / agent loop you've built (repo, demo, or detailed description) 2. Which ad platform API you've worked with and what you shipped on it 3. How you'd handle Meta's rate limits across many client accounts — short paragraph is fine Proposals that don't address these will be skipped. I'd rather see a short focused reply than a templated pitch.