Lead Data Scientist - Canada - Contract

(Remote – Canada)
About Very
Very is a fully distributed technology firm led by expert problem-solvers who create efficient, scalable solutions that move commercial, industrial, and consumer products from pilot to production in record time.
We believe that real innovation happens in the grind, working shoulder to shoulder with clients who are building the future. Our team thrives on that energy. When we're not helping clients deliver business-critical outcomes, we're refining our craft and celebrating what it means to do hard things well.
We've built a collaborative, tight-knit culture that thrives in both remote and in-person settings. We've won numerous workplace awards over the years, including Great Place to Work certification and recognition from Parity.org as a Best Company for Women to Advance.
Our clients include well-known brands like Vizio, Peloton, Clear, iHeart Radio, and Fellowes, all determined to leverage connected devices and AI to drive meaningful impact. Our job is simple: help them win.
About This Role
A Lead at Very is an individual who operates with the highest degree of knowledge and accountability for the delivery of services to our customers. They provide excellent technical leadership and delivery skills, as it pertains to complex, multi-faceted projects at Very. They have a strong executive presence, which gives major client stakeholders the confidence that we will deliver, and gives our team the confidence and accountability to do so.
As a Lead Data Scientist at Very, you own the data and modeling workstream on client engagements. You will be handed a vague client outcome and a small budget, and expected to independently decide the architecture, the evaluation design, and what to tell the client. You make architecture calls, defend them internally, present results directly to non-technical owners, and recognize when the highest-leverage next step is a change to the data or approach rather than further tuning.
The work is roughly 30 percent modeling, 30 percent data and infrastructure plumbing, and 40 percent client-facing judgment. This is not a research role. The people who succeed here are as comfortable owning a data pipeline end to end as they are training a model.
Lead engineers also serve as solutions engineers for the commercial team, helping close contracts with terms that are conducive to successful delivery.
This is not an easy role. You'll work in complex domains, under real deadlines, and with clients who expect you to bring clarity, confidence, and results. If you find satisfaction in doing hard things well, in solving tough problems, building real systems, and helping others rise to the challenge, you'll fit right in.
As a client services organization, travel may be required up to 10% of the time.
What You'll Be Working On
Almost all of our projects are production systems. Recent engagements have included two representative shapes of work:
Applied computer vision on physical-world measurement problems. Turning raw video into labeled training datasets, training and evaluating vision models against noisy real-world ground truth, standing up cost-aware hosted inference for large models, and translating error metrics into plain business language for a client who thinks in dollars, not RMSE.
Agentic data platforms. Ingesting unreliable public or client data into a well-designed relational schema, building hybrid lexical and semantic retrieval, exposing typed tool interfaces to LLMs, and defending the correctness of every answer to technically curious stakeholders who test the system adversarially.
These two shapes are representative, not exhaustive; the exact nature of your projects will vary. Across our engagements, we typically leverage the following:
Core: Python at a production engineering level, PyTorch, the SciPy stack, Git and GitHub Actions, agentic AI development (MCP servers, LLM APIs, typed tool design)
Data and backend: PostgreSQL including full text search and pgvector, Python web frameworks such as Django or FastAPI, Docker, and Celery and Redis for job orchestration that scales to zero between bursts
ML lifecycle: MLflow or Weights and Biases for experiment tracking, labeling platforms with API-driven upload, speech-to-text pipelines (Whisper), Jupyter for prototyping
AWS: SageMaker, Fargate, ECR/ECS, Lambda, RDS, S3, ElastiCache, IAM, Bedrock, IoT Core, Greengrass
Infrastructure-as-Code: Pulumi and Terraform
Edge: Embedded inference on NVIDIA Jetson, RPi, or ESP32; OpenCV; streaming pipelines
You will collaborate closely with our software, hardware, and design teams, so enough full-stack literacy to integrate a model into a running web application is expected.
We value well-tested, reusable code and expect our engineers to be as good practitioners as they are leaders and teachers.
Responsibilities
Work with clients, sales, engineers, and designers to define and estimate Statements of Work, including assumptions, risks, and dependencies
Own the data science components of an engagement end to end, from architecture through production and monitoring
Design the evaluation, not just report a metric: validation strategies that reflect how the system will actually be used, metrics appropriate to the problem, and for search systems, a labeled set of test questions with known answers so retrieval precision and recall can be measured and regressions caught
Own the correctness story of the systems you build: auditability, reproducibility, reconciliation against source, and traceability from any answer back to the underlying record
Architect, build, and deploy reliable data, retrieval, and ML pipelines into production
Present model and system results directly to non-technical clients in plain language, without jargon or over-promising
Push back on a bad idea from the client and offer a better path at the same time
Scope and estimate your own work in hours, and hold to that time box
Raise uncomfortable findings early, including when better data quantity or quality, or a different strategy altogether, is required
Write clearly: short summaries, model reports, schema documentation, and repos another engineer can run
Mentor senior engineers and review others' work, catching flawed experimental design and training configurations before they reach production
Establish and enact DataOps and MLOps best practices, and continue to evolve the Data Solutions practice at Very
Minimum Qualifications
Education
Master's degree in Data Science, Computer Science, or a related quantitative field. A PhD or equivalent research training is a strong signal but not required
Experience
5+ years of related experience, including a shipped model or data system that real users depend on
Python at an expert engineering level: production services, packaging, testing, code review, and the discipline to hand work to another engineer
Applied computer vision end to end: modern vision architectures (transformer-based and CNN backbones), a range of task heads, and transfer learning mechanics (fine-tuning strategies, learning rate schedules, regularization, early stopping)
Evaluation and experimental rigor: train/validation/test methodology that avoids leakage, diagnosing systematic error patterns, justifying metric choice for the problem at hand, and disciplined experiment tracking
Deep PostgreSQL: schema design for a messy real-world domain, indexing and query performance, full text search, and pgvector
Data engineering over unreliable sources: bulk acquisition, normalization, idempotent and incremental loads, and reconciliation against a source you do not control
Dataset construction, not just consumption: building datasets from raw source material, preserving traceability back to source, working with labeling platforms, and joining outputs to messy real-world ground truth
Retrieval systems, lexical and semantic: chunking strategy, embedding selection, hybrid retrieval and rank fusion
LLM application engineering: MCP or equivalent tool interfaces with typed and validated parameters, grounding, citation of underlying records, and hallucination control
AWS as a delivery target: SageMaker training jobs, GPU instance selection and cost awareness, Fargate, RDS, S3, and hosted inference endpoints that a live application calls
Infrastructure-as-Code (Pulumi or Terraform) and GitHub Actions CI/CD
Led cross-functional engineering teams, and partnered with sales and client success to secure and grow work
Strong written and spoken communication skills in English
Nice to Haves
7+ years of related experience
Azure experience: Functions, Container Registry/Instances, SQL Database, Machine Learning, IoT Hub
Classical CV and object detection alongside deep learning
Multi-frame, temporal, video, or pose estimation models
Managed authentication and JWT validation patterns
Prior consulting or agency delivery experience with fixed-price, fixed-scope work
AWS Professional level certification
Requirements
Must reside in Canada
Must be legally authorized to operate as an independent contractor in Canada
Skills
In addition to experience, these are the critical skills we look for in all technical roles, and how they should be demonstrated at the Lead leve

Work arrangement
Yes