SENIOR PYTHON / AI AGENT ENGINEER

SENIOR PYTHON / AI AGENT ENGINEER

HeartStamp | Remote | Full-time contract | Start immediately

Own the agent that sits at the center of a live consumer product.

ABOUT HEARTSTAMP

HeartStamp is an AI-native platform that helps people create one-of-a-kind greeting cards, professionally printed and mailed to the door, or delivered as an animated digital card. We launched in the US in 2026 and we are shipping fast: digital 3D cards, direct mail, invitations, and a mobile app.

We are built for Millennials and Gen Z who find traditional cards generic and emotionally hollow. A customer talks to Stampy, our conversational assistant, and comes away with a card that looks like the one they would have made if they had the time, the talent, and the taste.

We are a small, high-output distributed team. Every engineer here owns a real surface and ships to real customers.

THE ROLE

Stampy is not a chatbot bolted onto a storefront. It is a stateful, tool-using agent that captures what a customer means, resolves it against a live catalogue of tens of thousands of cards, drives image generation, and hands off into a print-ready pipeline. It is the product.

We are hiring a Senior Python / AI Agent Engineer to work on that system with a clear path to owning it end to end: the agent graph, the retrieval layer underneath it, and the deterministic control layer around it.

That last part is the job. The hard problems here are not "can the model answer." They are: does the captured intent survive the handoff, does the tool get invoked on the right slot, does the state machine recover when the model drifts, and can you prove any of it with an evaluation suite rather than a vibe check. If you have shipped agents into production, you already know that the failures that hurt are the confident wrong answers, not the crashes.

You will work directly with our Tech Lead, Principal Engineer and Founder, and you will own decisions rather than tickets.

THE HARD PART

Read this section carefully, because it is the real job and it is what we will talk about if we speak.

A conversational agent and a product catalogue speak different languages. The agent can express far more than any storefront has pages for, and the translation between those two vocabularies is where agent products quietly break. Not with errors. With plausible answers that are wrong.

The layer that turns what a customer means into catalogue state, and keeps both vocabularies honest with each other, is the highest-value thing you will own here. Getting it right is mostly not a model problem. It is a contract problem, a state problem, and an evaluation problem.

If that sounds interesting rather than tedious, we should talk.

WHAT YOU WILL OWN

The agent layer

• Stateful, tool-using LLM workflows in LangChain and LangGraph: structured outputs, routing, checkpoints and memory, approval gates, multi-step behavior.
  • The deterministic control layer: reliable slot and state capture, stage gating, guarding tool invocation, and graceful recovery when the model drifts.
  • Multi-provider integration (OpenAI, Google Gemini, Vertex AI, OpenRouter, xAI) with model selection, prompt versioning, rate limits and failure handling. We are deliberately model-agnostic.
  • Interactive performance: per-turn latency budgets, response streaming, prompt caching, model routing for sub-tasks, and the cost and speed trade-offs behind each.

Retrieval

• pgvector, embeddings, HNSW indexing, and the full RAG pipeline: ingestion, chunking and metadata design, hybrid retrieval, ranking, relevance tuning, fallback strategies, and retrieval evaluation.
  • Retrieval that serves a real catalogue with real inventory, not a document Q and A demo.

The resolution layer

• The mapping between conversational intent and catalogue state: occasion slugs, recipient and filter tokens, style tokens, and the routes they resolve to.
  • Keeping the agent's vocabulary and the storefront's vocabulary in sync, and making that a checked contract rather than a runtime guess.
  • Round-trip guarantees: if we put a value in a URL, reading it back yields the choice that produced it.

Proving it works

• pytest, async tests, mocked external dependencies, and agent-specific evaluation suites.
  • Golden datasets, automated behavioral tests, retrieval metrics, LLM-as-a-judge, and production feedback loops.
  • Structured logging, tracing, error monitoring, and LLM tracing with LangSmith.

You will also work in

• FastAPI: async APIs, Pydantic, REST design, auth, error handling, integration testing.
  • PostgreSQL with SQLAlchemy or SQLModel, and query performance.
  • Celery and Redis for background AI pipelines: retries, idempotency, backfills, and long-running jobs. We have people who own the infrastructure, so you do not need to be deep here, but you need to be comfortable.

MUST-HAVE QUALIFICATIONS

• 5+ years of professional Python engineering, including ownership of production services.
• 2+ years building LLM-powered features, agentic workflows, semantic search, or RAG systems that real customers used.
• Demonstrable experience shipping retrieval backed by PostgreSQL and pgvector, beyond prototype chatbots.
• Solid FastAPI experience inside a distributed system of APIs, workers, queues and third-party services.
• Real evaluation practice. You can describe how you measured whether an agent got better or worse, and what you did when the number moved the wrong way.
• Pragmatic reliability judgment around provider outages, model failures, retries, graceful degradation, and feature-flagged rollouts.
• Comfortable with Docker, CI/CD, and AWS.
• Strong written English. You will be arguing about architecture in Slack with senior engineers and writing design notes people act on.
• Minimum 4 hours of daily overlap with 9am to 5pm US Eastern Time.

NICE TO HAVE

• Enough TypeScript to follow a bug across the boundary. Some of our worst defects live half in Python and half in the frontend's filter state. You do not need to be a React engineer, but you need to be able to read it.
• Multimodal AI, especially image generation or image understanding pipelines. Strongly preferred.
• Evaluation tooling: DeepEval, LangSmith evaluations, Sentry, PostHog.
• OpenAI-compatible APIs, OpenRouter, xAI/Grok, Gemini, Vertex AI.
• S3-compatible storage, image moderation, and content safety.
• A consumer product where AI output directly drives a purchase or a creative result.
• Prompt versioning, and evaluating conversational quality across multi-turn flows.

WHAT THIS ROLE IS NOT

• Not a prompt engineering role. We have a prompt engineer and a card library team.
• Not a research role. Everything you build goes to customers this quarter.
• Not a greenfield rebuild. The platform is live and shipping. You are strengthening and extending a real system.
• Not an infrastructure role. We have a DevOps engineer and a principal engineer who own that.

ENGAGEMENT DETAILS

Type: Full-time independent contractor, approximately 50 hrs/week. Overseas hires on a contractor agreement.

Term: 90-day initial contract with a strong path to extension based on results.

Compensation: $2,000 to $3,500 USD per month, weighted to demonstrated agent and retrieval depth. We are stating this openly so nobody wastes their time.

Location: Remote. Pakistan, the Philippines, and the wider South and Southeast Asia region.

Hours: Minimum 4 hours daily overlap with 9am to 5pm US Eastern Time.

Reports to: Tech Lead, with direct access to the Founder and Principal Engineer.

Start: Immediately.

SCREENING QUESTIONS

Please answer these directly in your application. Substance matters more than length, and short specific answers beat long general ones. Applications without answers will not be reviewed.

1. An agent captures a user's intent correctly, the parameter is present and correct in the request, and the user still lands on the wrong result. Nothing errors and every test passes. Walk me through how you find that, and what check you add so it cannot happen again.

2. Describe an evaluation suite you built for an agent or a retrieval system. What did it assert, how often did it run, and what did it catch that a unit test would not have?

3. You have a multi-step conversational flow where the model sometimes skips a required slot. How do you make the flow reliable without making it feel scripted?

4. Tell me about a retrieval system you shipped on PostgreSQL and pgvector. What was the corpus, how did you chunk it, and how did you know your ranking was any good?

5. What are your current working hours, and what overlap can you commit to with 9am to 5pm US Eastern?

HOW TO APPLY

Put your answer to Question 1 at the very top of your message. Then send your resume or profile, a link to production work you can speak about in detail, and your answers to the remaining questions.

We review daily and move fast. Hiring THIS WEEK. BY FRIDAY SEPT 4