Who We Are
At Bennett Data Science, we've been pioneering the use of AI, predictive analytics, and data science for over a decade for some of the biggest brands and retailers. We're at the top of our field because we focus on delivering actionable AI for our clients. Our deep experience and product-first attitude set us apart from other groups and gets us the business results our long-term clients want.
Why You Should Work With Us
You'll be exposed to a wide range of clients who are at the cutting edge of innovation in their field and get to work on fascinating problems, supporting real products, with real data. We help lots of companies, from some of the largest companies in the world to small startups in Silicon Valley who are building the next big thing.
About
We are hiring for a remote applied AI consulting firm that builds production AI systems for clients. The company works on real business workflows, private data, and deployed software. The team is small, senior, and focused on systems that survive production use. The company values clear ownership, strong engineering judgment, and direct communication.
The role
You will own a reusable agent platform that powers every agentic client engagement. The platform will cover orchestration, tool calling, retrieval, evaluation, deployment, monitoring, and common data connectors. You will take the current foundation forward and make it stable, documented, and easy for other engineers to use. You will make key architecture choices and keep the system simple enough to extend across different client environments. You will work closely with senior engineers who adapt the platform to client workflows.
You will:
deliver a scalable agentic platform from early prototypes to a deployable, sellable product
build the core services that let agents plan work, use tools, retrieve data, and complete tasks safely
apply proven practices for reliability, security, monitoring, cost control, and recovery
create clear ways to test quality, review production behavior, and improve the platform based on real usage
make the platform easy to deploy, operate, and adopt across client environments
write clear guidance for engineers who build on the platform.
Required:
LangGraph in production is essential. You have designed, shipped, and maintained LangGraph graphs in production: state management, checkpointing, multi-agent orchestration, and human-in-the-loop interrupts
7+ years shipping production software, with at least 2 years on LLM-powered systems running in production (not an internal tool)
Strong Python. Comfortable with FastAPI or equivalent for serving, Postgres or equivalent for state, and Docker for deployment
Built or maintained a non-trivial agent system end to end: orchestration, tool calling, retrieval, evaluation, deployment
RAG experience: hybrid retrieval, chunking strategy, evaluation, citation grounding, no-answer fallback
Production cloud deployment on AWS, GCP, or Azure. Comfortable with continuous integration and deployment, observability, on-call
Evaluation discipline: you measure on production traffic, you write the eval harness, you do not ship without it
Written and spoken English at B2 or above
Can take an ambiguous problem and ship a working system without daily supervision
Strong plus:
Open-source contributions to LangGraph or LangChain, or you have published LangGraph patterns others use
You have built a platform or framework that other engineers use. You have written the documentation for it
Comfort with a second orchestration approach beyond LangGraph (Claude Agent SDK, AutoGen, CrewAI, or hand-rolled) so you can make the right call per client
Fine-tuning experience: LoRA, PEFT, instruction tuning, distillation
Open-source contributions to LlamaIndex, DSPy, Haystack, or similar
Background in classical ML (scikit-learn, gradient boosting, time series)
MCP (Model Context Protocol) experience or interest
Shopify ecosystem, Snowf
