About the job - AI Architect ContactMonkey's platform already runs AI in production - Claude-powered template editing, a standalone ConfidenceCheck service that evaluates outgoing emails for editorial quality and link safety, Braintrust-backed evaluations, and per-organization feature gating. The next wave is bigger: grounded assistants, retrieval over customer brand and campaign history, richer agentic flows inside the composer, and data products that make sending decisions smarter.
We're forming a dedicated AI pod to own that trajectory. With the team you will be setting the technical direction for the pod, make the foundational infrastructure bets (vector store, evals, data warehouse, agent orchestration), and ship the first production systems yourself. This is an architect role in the sense that you own the shape of the thing - it is not a role where you stop writing code.
Your impact- AI platform direction: Model selection and routing, prompt and context architecture, evaluation strategy, cost and latency budgets, safety and PII handling. Decide where we go deeper with frontier models, where we fine-tune smaller ones, and where classical ML beats an LLM.
- Retrieval and grounding infrastructure: You'll pick it, design the ingestion path (brand kits, past campaigns, template library, per-org knowledge), own the chunking/embedding/refresh strategy, and set the quality bar via evals.
- Agent and tool-use patterns: Multi-turn tool-calling loops, deterministic fallbacks, observability, recoverable failure modes across Sidekiq and SQS, human-in-the-loop checkpoints where they matter.
- Data engineering foundations: You'll decide whether we need a warehouse (BigQuery / Snowflake / DuckDB-on-S3/ RedShift), design the pipelines that feed both product analytics and model training/eval datasets, and make sure the lineage is clean enough that we trust what we ship. Build a foundation for the next level data analytics across the company.
- Cloud and deployment architecture: We run on AWS (ECS, S3, SQS etc). You'll make the calls on how AI services are deployed - when a new microservice like ConfidenceCheck is the right pattern vs. embedding in the Rails monolith, how we handle GPU/accelerated inference if we ever need it, how we isolate model spend per tenant.
What we expect from you technicallyYou should be credible across all four of AI, Cloud, Data, and Fullstack - not a specialist in one with passing knowledge of the others.
Concretely:
AI / ML- Built and operated LLM-backed features in production against real users.
- Deep familiarity with at least one major provider SDK (Anthropic, OpenAI) and comfort switching between them. Understand prompt caching, structured outputs, tool use, and the tradeoffs between agent loops and pipeline-style orchestration.
- Evals are a reflex, not an afterthought. You've run a real eval harness (Braintrust, promptfoo, in-house) against golden datasets and used it to block regressions.
- Working knowledge of retrieval systems: embeddings, hybrid search, reranking, and where each breaks. Opinions on pgvector vs. dedicated vector DBs and when either is wrong.
- Bonus: fine-tuning or distillation experience, classical ML for ranking/classification, prompt-injection and jailbreak threat modelling.
Cloud- AWS fluency at the architecture level - VPCs, IAM boundaries, secrets management, queue-based decoupling, cost control. Equally comfortable if we need to run a workload on GCP for a specific reason.
- Container-native deployment (ECS, EKS, or equivalent). IaC experience (Terraform or CDK) - even if we aren't using it yet, you should expect to bring it in.
- Have operated production systems under real load. You've been paged, and you've written the postmortem.
Data Engineering- Can design a data pipeline end-to-end: ingestion, transformation, storage, serving. Batch and streaming. Have built this in at least one real organization.
- Comfortable with SQL and non SQL at a level that lets you reason about query plans, not just write queries. You've tuned a slow query by looking at the plan, not by guessing.
- Opinion on when a warehouse earns its keep vs. when read replicas are enough. We're at that decision point.
- Bonus: dbt, Airflow / Dagster / Prefect, CDC patterns, multi-tenant data partitioning at scale.
Fullstack- Our backend is Ruby 3.4 on Rails 7.2 with Sidekiq Pro and SQS; our frontends are Vue 3 (main app), Nuxt 4 (recipient-facing content), and a Fastify/TypeScript AI microservice. You don't need all of this on day one, but you should be a polyglot who ramps quickly.
- Write real TypeScript. The AI-adjacent services are TS-first and that is likely to grow.
- Understand API design past the happy path - versioning, auth modes (we run session auth for the main API and bearer tokens for integrations), rate limiting, idempotency.
What we bring to the table 100% employer-paid benefits + a Health Spending Account from day one
Work from anywhere in the world for up to 4 weeks
Stock option plan-own a piece of our success
RRSP Group Savings Plan to plan for your future
Generous vacation package to recharge and relax
Personal development budget to fuel your growth
One personal day + two volunteering days to give back
Your Birthday off-celebrate on us!
Five health days per year to stay at your best
Beautiful downtown Toronto office for hybrid work-fully stocked with all the best snacks
Compensation & Work DetailsCompensation is thoughtfully determined based on your experience, skill set, and alignment with our internal compensation framework and internal equity.
We're always happy to answer questions about compensation throughout the hiring process.
This is a net new position based out of our downtown Toronto office, at King and Spadina. Our engineering team works primarily remote but we are looking for someone who is willing to come into the office occasionally on an as needed basis.
