AI Agent Infrastructure Engineer (Python)

We're building scalable, multi-tenant agentic infrastructure — autonomous AI agents that connect to external APIs, execute real actions, and serve multiple users with strict isolation. This is not a chatbot wrapper. This is production agent infrastructure with tool execution, credential management, sandboxed compute, and real-time streaming.

You'll own the full stack — from the Python backend and agent runtime to the Next.js frontend.

---

WHAT YOU'LL BUILD

- AI agent runtime — Multi-turn reasoning with tool-use via Anthropic Claude Python SDK. No LangChain. Raw agent loop.
- Custom MCP servers — Python tool servers wrapping external APIs (ad platforms, email, productivity tools) using the MCP protocol.
- Multi-tenant backend — FastAPI service with per-tenant OAuth credential management, encrypted storage, and strict data isolation.
- Sandboxed execution — Docker-based isolated Python environments for per-tenant code execution, data processing, and report generation.
- Real-time streaming — WebSocket-based token streaming with tool call interleaving (FastAPI ? Next.js).
- Automation engine — Celery-based trigger system for scheduled tasks, alerts, and webhook-driven agent actions.
- Billing & usage metering — Stripe integration with tiered plans and per-tenant usage tracking.

---

TECH STACK

Python 3.12 | FastAPI | SQLAlchemy | Celery | Pydantic | Next.js | TypeScript | Tailwind | Anthropic Claude | MCP Protocol | PostgreSQL | Redis | Docker | Railway | Stripe

---

MUST-HAVE REQUIREMENTS

AI Agents
Built and shipped AI agents with tool-use in production. Understands ReAct loop, tool routing, context management. No frameworks — raw SDK only.

Python
Expert-level: async/await, FastAPI, SQLAlchemy 2.0, Pydantic, Celery.

LLM API
Anthropic Claude Python SDK — Messages API, tool_use, streaming, multi-turn.

OAuth & Security
OAuth 2.0 implementation in Python. Credential encryption. Secure token lifecycle management.

Multi-Tenant Architecture
Built systems with strict per-tenant data isolation, scoped queries, and tenant-aware middleware.

API Integrations
Experience integrating complex external APIs (ad platforms, CRMs, or similar) with pagination, rate limits, and error handling.

Frontend
Next.js / React / TypeScript — can build a polished chat UI, dashboards, and onboarding flows.

Real-Time
WebSocket or SSE streaming between a Python backend and React frontend.

---

NICE TO HAVE

- MCP (Model Context Protocol) server development
- Docker container management and sandbox isolation
- Celery + Redis for distributed task queues and scheduling
- Stripe billing (subscriptions, usage-based metering)
- pandas / matplotlib for data processing and visualization
- Google Ads API or Meta Ads API experience
- Prior startup / 0?1 product experience

---

WHAT MAKES THIS DIFFERENT

You're not wiring up a chatbot or plugging into a pre-built framework. You're architecting the agent runtime itself — tool routing, MCP server registry, per-tenant credential injection, sandboxed execution, and real-time streaming. If you've wanted to build agent infrastructure from the ground up, this is it.

---

DO NOT APPLY IF:

- You plan to use LangChain or LlamaIndex — we need raw SDK understanding, not framework abstraction.
- You want to build the backend in Node/TypeScript — the backend is Python, non-negotiable.
- Your experience is chatbots only — chatbot ? agent. We need tool-use with real external actions.
- You have no async Python experience — the entire backend is async FastAPI.
- You have no OAuth experience — credential management is the critical path.
- Your Python is weak — this is 80% Python work.
```

---

---

**JOB TYPE**
```
Full-Time / Part-Time (select both if option available)
```
Python, FastAPI, PostgreSQL, Docker, Redis, Next.js, TypeScript, WebSockets, Stripe, AI/ML, REST API
```