Senior AI Engineer

Senior AI Engineer

Ecolab Digital is seeking a Senior AI Engineer to design and ship the production LLM and agent systems that turn data into intelligent, customer-facing intelligence products across the Institutional & Specialty segment. This is a builder role. You own systems end to end, from the data layer through inference, serving, and product integration, and you write code every day.

You will work across LLM-based applications and multi-agent workflows, combining GenAI reasoning with deterministic logic to produce outputs that are reliable, auditable, and feasible to run at scale. Partnering with product, data science, and engineering teams, you will take systems from first prototype through production and the long tail of operating them: evaluation, monitoring, cost, and latency. You care about the craft of engineering as much as the model.

What you will do

• •Design, build, and ship production LLM applications and agentic systems end to end, owning the technical decisions from data through inference, serving, API exposure, and product integration.
• •Build multi-agent and multi-step workflows: tool-calling, sequential handoffs, and scheduled or DAG-based orchestration, with sound judgment on when to use probabilistic reasoning versus deterministic rules.
• •Treat prompt and context engineering as a real discipline: structured prompts, output schemas, exclusion rules, and deliberate management of what goes into the model's context window.
• •Build retrieval-augmented generation over vector indexes: chunking, embeddings, hybrid search, reranking, and evaluation of retrieval quality, not just of the final answer.
• •Own the production lifecycle of the systems you build: evaluation harnesses, monitoring and tracing, drift detection, prompt and version management, champion or challenger promotion, and retraining.
• •Engineer for cost and latency as first-class requirements: caching, model routing, batching, and streaming, and sound judgment on when to spend inference on a reasoning model versus route to a faster, cheaper one.
• •Build in guardrails and security from the start: prompt-injection defense including through retrieved content, protection against tool misuse and data exfiltration in agentic systems, PII handling, audit logging, and authentication on anything customer-facing.
• •Expose AI capabilities as production services and APIs, with attention to latency, reliability, versioning, and auth.
• •Partner with product, data science, and engineering to translate requirements into systems that scale, and communicate design, tradeoffs, and results clearly to technical and non-technical audiences.

Minimum Qualifications

• •Bachelor's degree in Computer Science, Data Science, Math, Statistics, or a related field with 5 years of AI, ML, or software engineering experience, or a Master's degree in a related field with 3 or more years.
• •4+ years of Python. You write clean, modular, production-grade code and treat version control, testing, and code review as standard practice, not afterthoughts.
• •2+ years of SQL for querying and preparing data.
• •Hands-on experience with PySpark and DataFrame APIs on a large-scale distributed platform, comfortable moving from a notebook to production-scale data.
• •3+ years building machine learning models, covering training, tuning, and evaluation. Solid ML and statistics are how you keep GenAI systems honest, not something GenAI replaces.
• •2+ years shipping production LLM applications. Prompt engineering as a discipline, RAG over vector indexes, and GenAI reasoning combined with deterministic logic for reliable, auditable outputs. You should be able to show what you shipped, who used it, and how you improved it after launch.
• •2+ years hands-on with GenAI frameworks and tools: LangChain or equivalent, Anthropic, OpenAI or Hugging Face APIs, and vector databases.
• •1+ year building and orchestrating multi-agent or multi-step systems: tool-calling, sequential handoffs, and scheduled or DAG-based workflows.
• •Experience exposing AI as production services and APIs, with attention to latency, reliability, versioning, and auth. You can put a model behind a service other teams depend on, not just run it in a notebook.
• •2+ years of production operations for AI or ML, with evaluation-driven development at the center. Evaluation, monitoring, tracing, cost, and latency. You build the eval set and regression gates before you tune, calibrate an LLM judge against human-labeled ground truth, and can prove a prompt or model change actually made things better rather than just confirming it still runs. You have operated real systems, not only proofs-of-concept.
• •Solid engineering and DevOps fundamentals as a normal part of how you ship: Git, CI/CD, containerization with Docker, and owning your services in production including logging, monitoring, secrets handling, and safe rollbacks.
• •Strong analytical thinking and communication, including product judgment in a probabilistic setting: knowing what "good enough" is and setting realistic expectations with stakeholders for systems that are not deterministic.

Preferred Qualifications

• •Experience with the Databricks GenAI stack: Model Serving, Unity Catalog, Vector Search, MLflow including prompt registry and champion or challenger evaluation, and Databricks Asset Bundles.
• •Experience with Model Context Protocol (MCP) servers and tools, conversational AI assistants, and emerging agent-to-agent (A2A) orchestration patterns.
• •Experience with agent memory and state management for long-running or stateful agents.
• •Judgment on when to fine-tune or distill a model versus prompt or retrieve, and experience doing at least one in production.
• •Proficiency across the Microsoft Azure suite (App Service, Functions, Key Vault, ADO Pipelines), and infrastructure-as-code or Kubernetes for multi-environment deployment.
• •Fluency using AI-assisted development tools as a genuine productivity multiplier.

Sharing a public GitHub profile or project portfolio is encouraged; we would love to see examples of your hands-on work where available.