Overview
Build the AI-native backend services and evaluation loops that bring LLM-powered features into production to make visa processing faster and simpler for travellers.
What you'll do
- Build AI-native backend services in Python that orchestrate LLMs, agents and tool calls across core product workflows.
- Build evals and reinforcement learning loops that measure model quality on real workflows.
- Use RL techniques (reward modelling, RLHF/RLAIF, RL fine-tuning) to continuously improve agent performance.
- Scale for volume and cost with async, event-driven workflows that handle peak demand while keeping latency and token spend in check.
- Own systems end to end from design to production, with strong testing, monitoring, observability and evals.
What you'll need
- 4-7 years of backend engineering experience.
- Hands-on experience building and running LLM-powered features in production.
- Exceptionally strong in Python, including async Python.
- Shipped production services with FastAPI, Django or Flask.
- Hands-on experience with LLM APIs (OpenAI, Anthropic, Gemini) and the building blocks around them: prompt design, tool calling, structured outputs, RAG and vector databases (pgvector, Pinecone, Weaviate or similar).
- Knowledge of how to evaluate and harden AI systems, including eval datasets, LLM observability tools (Langfuse, LangSmith or similar) and trade-offs between accuracy, latency and cost.
- Solid fundamentals in databases, distributed systems and cloud infrastructure, including PostgreSQL, Redis, message queues (Kafka, SQS or Celery), AWS or GCP, Docker and Kubernetes.
Nice to have
- Exposure to agent frameworks (LangGraph, LlamaIndex or similar), fine-tuning, reinforcement learning (RLHF, reward modelling) or self-hosted models, browser automation (Playwright), or building in fintech, travel or other high-compliance domains.
Details
Read the full description and apply on the company’s own careers page.