Senior AI Engineer

Catapult is building the future of sports performance technology.

Since 2006, we’ve helped more than 5,000 teams use data, science and technology to improve athlete health, readiness and performance. Our customers include teams across the NFL, NBA, NHL, MLS, EPL, AFL, NRL, NCAA and many more.

Now we’re building the next layer of that platform: AI that can reason across everything we know about an athlete and turn it into intelligence a coach or performance practitioner can trust.

We’re looking for an Agentic AI Engineer who has already shipped production AI systems and understands what it takes to make them reliable, measurable and trustworthy.

What you’ll do:

• Design and ship specialist AI agents that use memory, tools, data and multi-step reasoning.
• Build multi-agent orchestration that routes work between specialist agents, manages dependencies and synthesises conflicting outputs.
• Develop systems that evaluate confidence, uncertainty and consequence before recommendations reach a practitioner.
• Build human-in-the-loop escalation so the system knows when to answer, when to ask for more information and when to defer to a human.
• Create workflows that turn sport scientist expertise into validated, versioned and testable agent capabilities.
• Build evaluation, observability and regression testing so agent performance can be measured and improved in production.
• Work with domain experts to ensure AI outputs are grounded, traceable and actionable.

The goal is simple: multiple specialist agents working together to answer complex performance questions with a recommendation that is fast, grounded and calibrated.

What you need:

This is a senior engineering role. Three technical capabilities are essential.

1. Production agentic AI

You have personally shipped a production agentic AI system used by real users.

You have hands-on experience with:

• Memory or persistent state
• Tool use or tool calling
• Multi-step reasoning or workflows
• Production deployment and operation

Chatbots, prompt engineering and RAG alone are not enough.

1. Multi-agent orchestration

You have built or substantially contributed to a production multi-agent system .

You understand:

• Agent routing and orchestration
• Specialist agent composition
• Dependency-aware workflows
• Parallel and sequential execution
• Conflicting agent outputs
• Response synthesis

Experience with LangGraph, AutoGen, CrewAI or equivalent frameworks is valuable.

1. Confidence calibration

You have hands-on experience calibrating probabilistic ML or AI systems.

You should be comfortable with:

• Platt scaling
• Isotonic regression
• Expected Calibration Error (ECE)
• Reliability and calibration curves
• Confidence and uncertainty estimation

We care about the difference between a model that sounds confident and a system with measurably calibrated confidence.

You should also have:

• 5+ years of professional experience in applied ML, AI or software engineering
• Strong Python
• Strong software engineering fundamentals
• Experience building and operating production systems

Experience with:

• Production RAG and reranking
• Foundation-model fine-tuning or domain adaptation
• LoRA, PEFT or similar techniques
• LLM observability and drift detection
• Evaluation harnesses and automated regression testing
• Human-in-the-loop architectures
• Confidence thresholds and escalation models
• Causal or counterfactual reasoning
• Go/Golang
• AWS, including ECS, EC2, Lambda, SNS or SQS
• GraphQL, REST or gRPC
• PostgreSQL or MongoDB

Experience working with sport scientists, clinicians or other domain experts is a plus.

You don’t need to be a sports scientist, but familiarity with workload, readiness, recovery, biomechanics or athlete performance data will help.

What success looks like:

You’ll help build a platform where specialist agents can investigate complex performance questions, use the right evidence, assess their uncertainty and produce a recommendation that a practitioner can understand and trust.

Most importantly, the system will know when not to answer.

Every recommendation should be:

Grounded. Calibrated. Traceable. Escalation-aware.

The practitioner remains responsible for the decision. Your job is to make that decision better informed, faster and more defensible.

Before you apply:

If you can demonstrate all three of the following, we’d like to hear from you:

1. Production agentic AI
2. **Product…