Senior Machine Learning Engineer

SELLIT9; Toronto; Hybrid; CAD 110,000 - 170,000; Full time

Job summary

SELLIT9 is seeking a data scientist to advance our ML-driven pricing and condition-recognition system. You will develop models that value used devices, grade condition from photos and videos, and push improvements into production with measurable impact.

You will lead end-to-end data work: dataset design, model evaluation, and collaboration with engineering to embed your models in the pricing engine. Strong Python and CV experience are essential.

Qualifications

• 5+ years in data science or ML
• Fluent in Python and ML stack: pandas, scikit-learn, and a DL framework
• Hands-on computer vision experience: image classification, object detection, or damage detection
• Strong statistical rigor; design evaluations and defend model behavior
• Experience with major LLM APIs: prompt engineering, multi-modal inputs
• Experience with agentic coding tools (Claude Code, Codex)

Responsibilities

• Build pricing predictor from measured gaps to suggested payout, graded against outcomes
• Lead condition grading end-to-end from photos/videos to pricing
• Build recognition models to identify category, model, specs from images/video
• Own the data: define damage taxonomy, design labeling operations
• Own model quality: evaluation frameworks, error analysis, retraining loop
• Use multi-modal LLMs as validation layer and for explanations
• Partner with engineering to ship models into the quote flow and pricing engine

Skills

Python

Machine learning

Data science

Computer vision

Statistical rigor

LLM APIs

Prompt engineering

Agentic coding tools

Tools

Pandas

scikit-learn

PyTorch

TensorFlow

BigQuery

PostgreSQL

Vertex AI

SageMaker

Docker

Job description

About the Role

SELLIT9 runs on one number: what a used device is worth. Two models decide whether we get that number right, and you'll own both.

The first is pricing. We're building an append-only, point-in-time-correct dataset of competitor payouts, retail prices, and resale comps, built specifically to be trained on. The deterministic version ships first: measured gaps against rivals, ranked by exposure. You'll take it from those rules to a price predictor, graded automatically against the price-move ledger we already run.

The second is condition and recognition. Today a seller tells us what shape their device is in and we verify it by hand on arrival. You'll build the computer vision that grades condition from photos and videos, and that recognizes what an item actually is: category, model, specs, damage.

This is a data science seat that has grown into production machine learning, not a software engineering seat that touches AI. We're hiring for the progression: analysis, then models, then models running in production with real consequences. You'll define the damage taxonomy, the labeling process, the evaluation bar, and the retraining loop, and you'll partner with the engineering team to ship what you build.

ML Tech Stack

We're early and open to the best tools for the job. You'll work across:

• Data foundation: PostgreSQL with monthly-partitioned market observations, the price-move ledger that grades every price change against a control group, and BigQuery.
• Modeling: Python end to end: pandas, scikit-learn, and a modern deep learning framework (PyTorch, TensorFlow, or equivalent), for tabular and vision models alike.
• Multi-modal: a major LLM API for validation layers and damage explanations: prompt engineering, structured outputs, vision inputs.
• Serving: cloud ML infrastructure (Vertex AI, SageMaker, or equivalent) and containerized model serving, monitored in production.
• How we build: coding agents across the team (Claude Code, Codex, and similar), a private library of skills and plugins, and MCP servers for our internal tools.

What You'll Do

• Build the price predictor: start from measured competitor gaps, end with a suggested payout per item, condition, and channel, graded automatically against real outcomes.
• Lead condition grading end to end: photos and videos in, a Flawless / Good / Used grade out, feeding straight into pricing.
• Build recognition models that identify category, model, and specs from images and video, so intake gets faster and less manual.
• Own the data: define the damage taxonomy, design the labeling operation, and keep the observation dataset model-ready.
• Own model quality: evaluation frameworks, error analysis, human review, and the retraining loop as labeled data grows.
• Use multi-modal LLMs as a validation layer for low-confidence cases and for customer-facing damage explanations.
• Partner with engineering to ship your models into the quote flow and the pricing engine.

Apply if you

• Have 5+ years in data science or machine learning, with the progression we're hiring for: from analysis, to models, to models you shipped and ran in production.
• Are fluent in Python and its ML stack: pandas, scikit-learn, and at least one deep learning framework.
• Have hands-on computer vision experience: image classification, object detection, or damage and defect detection.
• Have real statistical rigor: you design evaluations, know when a metric is lying, and can defend a model's behavior to a skeptical room.
• Have worked with major LLM APIs: prompt engineering, structured outputs, multi-modal inputs.
• Work AI-natively, and this is non-negotiable: agentic coding tools (Claude Code, Codex, or similar) are your daily driver. AI multiplies your craft, it doesn't replace it: you can defend every line an agent writes.

Even better if you have

• Pricing, forecasting, or demand-modeling experience where your model moved real money.
• Video understanding: temporal models, frame sampling, or quality assessment from video.
• Built a labeling and annotation operation from scratch, not just consumed someone else's dataset.
• Experience in recommerce, e-commerce, or marketplaces, where item condition drives the economics.