Build and improve production AI/data pipelines that run across LLMs, APIs, databases, and workflow systems like Temporal, Postgres, ClickHouse, and Kubernetes.
Design evals and build Jupyter notebooks that help us measure model behavior, data quality, extraction accuracy, and end-to-end customer impact.
Build observability into AI workflows so we can understand cost, latency, reliability, and quality.
Experiment with new models, retrieval strategies, structured-output techniques, prompt/program architectures, and model-routing approaches.
Integrate with fast-changing research and AI APIs, understand their behavior deeply, and build robust abstractions around them.
Help define the foundation for future inference serving, fine-tuning, dataset generation, and model evaluation infrastructure.
Requirements
A strong engineer who can reason through distributed systems, data pipelines, databases, and are comfortable working across varying programming languages.
You are extremely AI-fluent and actively use modern AI tools to move faster.
You have strong first-principles thinking and can turn ambiguous problems into hypotheses, experiments, and shipped systems.
You have good judgment and taste: you simplify aggressively, avoid unnecessary complexity, and care about maintainability.
You care about measurement. You do not trust vibes when evals, tests, traces, or data can tell you what is actually happening.
Bonus: You have experience with PyTorch, Hugging Face, vLLM, Ray, MLflow, RAG, fine-tuning & RL, inference serving, and/or model evaluation systems.
Tech Stack
Distributed Systems
Kubernetes
Postgres
PyTorch
Ray
Benefits
Massive upside: Meaningful equity in a venture-backed company defining a new category in commerce intelligence.
Real AI systems: You’ll work on production AI that directly affects customers, data quality, and revenue — not demos or toy agents.
Big surface area: You’ll help connect research ideas to production systems and customer-facing product.
Right moment: We’re past prototype, real brands already rely on us, and we have strong PMF — but the ceiling is still wide open.