AI Engineer

The role

OODARIS is the Agentic OS for Retail Decisions. Every night it finds what moved in a retailer's business, explains why and proposes a fix that a planner approves in the morning. ALDO Group backs OODARIS and runs it in production across ALDO, Sperry, Call It Spring and G.H. Bass.

We are early. OODARIS covers five merchandising decisions: merchandise financial planning and open-to-buy are in production at ALDO Group, and all five are available to new clients. You will work directly with the founder and our engineers.

You will own the agent layer: how agents plan, call tools and retrieve data, and how we evaluate and trace what they do. By default, nothing is written back to a retailer's systems until the decision's owner approves, and the code you write has to keep it that way.

The stack is asynchronous Python and FastAPI, Postgres with vector search, Redis and Docker, with OpenTelemetry tracing and a model layer that works across several LLM providers.

What you will do

• Build the multi-agent workflows behind each decision, with typed inputs and outputs for every tool call.
• Give each workflow a defined recovery path for when a tool fails, times out or returns bad data.
• Own retrieval over Postgres with vector search: chunking, indexing and hybrid search, each with targets for latency, quality and cost.
• Run the model layer across several LLM providers, with caching, routing and fallbacks.
• Build the evaluation suite: offline test sets, regression checks in CI and red-team cases that track grounding, tool success and cost per task.
• Trace every agent run from its input data to the fix it proposes, with structured logs and OpenTelemetry traces.
• Turn the questions planners ask OODIE, the agent they talk to, into evaluation cases.

Your first 90 days

• Day 30: a first change to a production agent workflow shipped, with an evaluation that runs in CI.
• Day 60: one agent workflow that you own end to end, with traces, alerts and a recovery path for every tool it calls.
• Day 90: a shipped improvement to that workflow in grounding, tool success, latency or cost per task, measured against the baseline you set at day 60.

What you bring

• 5+ years as a software engineer, including 2+ years shipping LLM or machine-learning systems to production.
• An LLM, agent or retrieval system you shipped that people relied on, and that you kept running after launch.
• Evaluations you built for an LLM system and used to decide whether a change shipped.
• Asynchronous Python services you built and ran in production, with FastAPI or a similar framework.
• A production incident you debugged from traces and logs, where you fixed the cause and not only the symptom.
• Clear writing: design notes that other engineers can build from.

It helps if you have evaluated and tuned retrieval with rerankers, fine-tuned models with adapters such as LoRA, or mapped product data across PIM and ERP systems.

Location

Montréal is preferred; remote in Canada also works, with core hours on Eastern Time. French is an asset, not a requirement.

Hiring process

1. A 30-minute intro call.
2. A technical conversation: walk us through a system you built in your strongest area (agents, retrieval, inference or evaluation), the trade-offs you made and how you knew it worked.
3. A practical session: work through a realistic agent or retrieval problem from our product with our engineers.
4. Conversations with the team, including product and engineering.
5. A final conversation with the founder, and reference calls with former managers.

How to apply

Use the form on this page, or email careers@oodaris.ai with "AI Engineer" in the subject line. Tell us about an LLM or agent system you built: what it did, how you measured whether it worked, and one decision you would make differently today.