Must-Have Skills:
- Strong software engineering fundamentals in at least one stack — clean APIs, sensible data models, tests, real debugging. Depth in one stack beats familiarity with five.
- Learning agility — a provable track record of entering an unfamiliar domain or codebase and shipping within weeks. This can substitute for most stack requirements.
- AI-native development with judgment — uses Claude Code or equivalent agentic tools as a primary way of building. Should be able to drive one live in the interview and explain where they took the keyboard back.
- High agency and builder's bias — takes an ambiguous ask and drives it to a shipped, measured fix.
- PostgreSQL competence — real queries, indexes, transactions, and the instinct that a wrong query is a customer-visible bug.
- Clear communication — can explain trade-offs to non-engineers without jargon, writes clearly, and is comfortable with real client contact.
- Working understanding of LLM agents — should be able to speak to what they've built with an LLM API and how it failed on them. Production-level depth is not required; we teach that.
Good-to-Have Skills:
Our environment (depth in some is a plus; candidate ramps on the rest):
- Python 3.11 (FastAPI, asyncio, pydantic)
- Anthropic and OpenAI SDKs
- Next.js / React / TypeScript at working proficiency
- Redis, Typesense
- AWS (EC2/ALB, RDS, SQS, SES)
- GitHub Actions, Datadog
Genuine bonus:
- Production LLM/agent work (evals, guardrails, multi-agent, MCP)
- Prior forward-deployed or consulting-style client work
- Airflow or data pipelines
- Browser automation, OCR/vision
Roles & Responsibilities:
- Build and operate AI agents and LLM-powered products — copilots, workflow agents, extraction pipelines, internal tools — and maintain/evolve existing ones.
- Own problems end to end: from the client conversation through design, code, deploy, and production support.
- Work AI-first, with humans owning judgment and outcomes — validate what the tools produce and own every line that ships.
- Take part in client and partner conversations: discovery, live demos of work in progress, pilot support.
- Write short proposals and design docs; decisions move through documents and PRs, not meeting chains.
- Run a lane end to end within a small, async-first team.
- Write tests where they protect what matters — agent behaviour and money paths. No coverage theatre.
- Apply AI security basics: prompt injection via untrusted content, least-privilege tool permissions, secrets hygiene, verifying anything inbound before acting on it.
