Senior Context Engineer

Responsibilities:
Build and own the system that captures everything known about each client's brand guidelines, design language, copy archives, past creative, and other textual assets and makes it queryable in milliseconds.
Build and own the system that pulls live data from each client's existing marketing stack campaign tools (e. g., Braze, Customer.io ), product analytics (e. g., Mixpanel, Amplitude), knowledge bases (e. g., Notion), communication channels (e. g., Slack), and behavioral data (e. g., Fathom) and normalizes it into context the agent can reason about. OAuth, rate limits, schema drift, and new integrations as the product grows.
Build and own the system that brings non-textual context into A images today, with video and other modalities to follow through embedding and retrieval choices appropriate to each modality.
Build the loop that lets A learn from itself, capturing what the agent did, what worked, what didn't, and the patterns across tool calls and user interactions, then feeding those learnings back as context that improves future runs.
Design the end-to-end context pipeline ingestion, chunking, vectorisation, storage, retrieval, and delivery with predictable latency, cost, and quality.
Build the evaluation framework that measures whether context is actually doing its job, retrieval-quality metrics, agent-quality regression tests, and dashboards the team uses to catch problems before users do. You ship this yourself; you do not delegate it.
Set the architectural direction for context across the platform patterns the team builds on, decisions documented in writing, and junior engineers contributing under your direction.
RFCs (Request for Comments) are end-to-end written design documents that specify the problem, alternatives considered, trade-offs, the decision, and what is explicitly not being built. Sharp enough that another engineer or an AI agent could build the right thing without you in the room.
Mentor a junior engineer or intern to ship-readiness on a scoped piece of the context surface.
Ship through AI agents. Write specifications Claude (or equivalent) can execute against, review the output critically, and ship the result. Engineers do not measure their value in lines of code typed.
Collaborate closely with the engineers who consume your context API, shaping what you produce to fit how the agent actually uses it.
Requirements:
5-7 years of total engineering experience, with at least 2-3 years on production context pipelines, retrieval-augmented systems, or retrieval at scale.
Has shipped a production retrieval or context system and run the evaluations that prove it works.
Both halves are required; we will ask for the specific evaluation design and what it caught.
Hands-on builder. You can sit down and ship the evaluation framework yourself this week, not delegate it.
Strong systems thinker. Opinions on vectorisation choices, chunking strategy, retrieval ranking, hybrid search, and cost/latency trade-offs and the ability to defend them.
Has built or owned at least two integration connectors end-to-end (OAuth, rate limits, normalisation).
Comfortable in our stack: TypeScript, Node.js, PostgreSQL, vector storage (pgvector or equivalent).
Strong adjacent stacks accepted if the ramp story is credible.
Written communication strong enough to author RFCs that another engineer or an AI agent can build from without you in the room.
AI-augmented engineering as a first-class part of your loop. If working through Claude (or equivalent) feels uncomfortable to you, this role is not a fit.
Smart and curious first-principles reasoning, comfortable in ambiguity, learns the next thing fast.

Experience
5-8 yrs