Our AI roadmap is live and shipping. You’ll work on systems that go from training and evaluation to production monitoring:
Survey timing optimization model — an ML system that learns the optimal moment to send a survey for each client relationship, maximizing response rates and data quality
NLP pipeline for free-text feedback analysis — classifying, scoring, and extracting structured signals from open-ended survey responses across thousands of enterprise clients
Client health scoring — an aggregate model that combines survey results, response patterns, historical sentiment, and relationship signals into a single predictive score per account
Agentic AI architecture using LLM orchestration (Google Vertex AI / ADK) — multi-agent systems that reason over client data and surface proactive recommendations to account managers
RAG system over enterprise knowledge bases — grounding LLM outputs in verified client data and platform knowledge
MLOps infrastructure: model versioning, A/B testing, inference cost monitoring, drift detection, and production observability for agent loops

Requirements

ML fundamentals you can derive, not just apply. You understand gradient descent, loss functions, regularization, and evaluation metrics at the level where you could implement them from scratch if you needed to.
Practical LLM experience. You’ve built something real with LLM APIs — function calling, structured outputs, context management, prompt design under constraints. You know how they fail and how to build around that.
NLP intuition. Tokenization, embeddings, semantic search, classification — you understand what’s happening inside the models you use.
Agentic architecture thinking. You’ve thought about or built systems where AI agents plan, use tools, and hand off to each other. You understand the failure modes: loops, hallucinated tool calls, context overflow.
Production ML mindset. You think about latency, cost, model drift, and monitoring before you think about accuracy metrics. A model that’s great in evaluation but unreliable in production is not a good model.
Python proficiency. You write clean, testable Python. You know when to use a dataclass vs a dict and why it matters at scale.
Bonus Points
Experience with Google Cloud AI stack: Vertex AI, Google ADK, Pub/Sub for agent communication
Multi-agent coordination patterns: orchestrator–worker, queue-based handoffs, tool use with guardrails
Fine-tuning experience — LoRA, PEFT, or full fine-tuning on domain-specific data
Java experience — our backend is Java/Spring Boot, and ML systems that integrate deeply with the platform need engineers who can cross that boundary
MCP (Model Context Protocol) integration experience
Experience with vector databases: Pinecone, Weaviate

Benefits

Why This Role Is Different
Most AI engineering jobs at this stage are either (a) prompt engineering wrapped in a Python script, or (b) infrastructure work with no meaningful ML. This role is neither. You’ll design learning systems, ship production models, and build agents that make decisions on behalf of enterprise clients. The data is real, the users are real, and the problems are genuinely unsolved.
We’re early enough that you’ll shape the architecture. We’re scaled enough that your work will be used immediately.
How We Hire
We hire on ability, not tenure. We don’t care whether your experience comes from a top university, a bootcamp, an open-source project, or a side hustle you built at 2am. What we care about is whether you can think clearly, build well, and learn fast.
Our interview process is deliberately hard. If you make it through, you’ll know you earned it — and so will we. We test fundamentals, systems thinking, and the ability to reason through problems you haven’t seen before. We don’t ask you to recite design patterns. We ask you to think.
Our AI/ML interview tests: ML fundamentals (you’ll derive things, not recite them), LLM system design, agentic architecture reasoning, and a practical exercise around a real problem from our domain. We care about how you think about failure modes, cost, and production reliability — not whether you can name every transformer variant. Strong Python and system design are expected. Java familiarity is a plus.

Applicant Tracking System Keywords

Tip: use these terms in your resume and cover letter to boost ATS matches.

Hard Skills & Tools

machine learning fundamentalsgradient descentloss functionsregularizationevaluation metricsLLM experienceNLPtokenizationembeddingsclassification

Soft Skills

production ML mindsetagentic architecture thinking