Software Architect — PageMind
Location: Barcelona / remote (EU time zones) · Type: Full-time · Reports to: CEO · Level: Senior / Principal
About PageMind
PageMind is the AI Catalog Intelligence platform for e-commerce. We turn product catalogs into structured, verifiable knowledge, enrich it with demand signals and expert knowledge, and generate the content that answer engines, search engines, and conversational assistants actually use — deployed directly into Shopify, WooCommerce, and BigCommerce stores. We work with retailers running catalogs of hundreds of thousands of SKUs, and with telco and enterprise clients on AEO content programs.
We are also an unusual engineering organization: most of our code is written by a fleet of AI coding agents (~130 of them) coordinated through infrastructure we built ourselves. The humans on the team design the systems, define what "done" means, own the hard decisions, and keep the fleet honest. This role sits at the center of that.
The role
You will own the technical architecture of PageMind: the product platform (ingestion, enrichment, knowledge graph, generation, and delivery to commerce platforms) and the engineering system that builds it (the agent fleet, its coordination layer, and the path from work item to production).
This is a hands-on architect role. You will write design documents and reference implementations, review what the fleet produces, decide what gets merged, and be accountable for the system behaving in production the way it was designed to. You will be the single most senior technical voice in the company after the CEO, and the person who says "no" when an agent — or a human — proposes rebuilding something that already works.
What you will do
Own the platform architecture
- Define and evolve the end-to-end architecture: catalog ingestion, attribute enrichment, the knowledge graph and signal databases, the generation pipeline, and write-back to Shopify, WooCommerce, and BigCommerce.
- Design for scale (catalogs of 10^5–10^6 SKUs, per-product traceability of every generation), for data quality, and for full cost attribution and auditability per customer and per job.
- Make the build-vs-buy, model-routing, and hosting decisions (multi-vendor LLM routing across Anthropic, OpenAI, and open-weight models; AWS Bedrock; self-hosted infrastructure) with explicit cost and risk tradeoffs.
- Own the schema. Missions and agents that take entity decisions on an old schema harm the architecture; you are the guardian against that.
Architect the engineering system itself
- Own the agent-fleet architecture: the pm-coord coordination layer (Cloudflare Workers + D1), mission and council servers, orchestrator behavior, and cross-vendor model routing (execution vs. architect vs. reviewer tiers).
- Define the definition-of-done and governance rules that let agents carry work safely from a work item to a production pull request: what gets tested locally before it reaches a shared branch, what a reviewer verdict must contain, what binds running processes to merged code.
- Fix the enforcement and observability gaps: long-running daemons that never reload merged code, verdicts that are lost on missed relay, workspaces that silently drift from main, claims that lapse without anyone noticing.
- Make agent output legible: reports that contain data a human can act on, not prose. Give the CEO real visibility of progress against the projects that matter.
- Design for the fleet to learn: mechanisms so a mistake fixed once is not re-implemented by the next agent working from a stale worktree or an outdated vault.
Ship and be accountable
- Write ADRs and design docs that agents and humans can both execute against. Review and approve next→main merges. Be on the hook for production.
- Lead the quarter's platform objectives: Shopify deployment, automatic onboarding of categories and customers, data-quality gates, and full cost attribution.
- Work directly with customers' technical teams when integrations or data models need to be negotiated.
- Mentor the platform developers and set the technical bar for contractors and future hires.
What we are looking for
Must have
- 10+ years building production software, with at least 3 years in a role where you were accountable for the architecture of a system others built on.
- Deep experience with distributed systems, event-driven and queue-based architectures, and data-intensive backends (relational and graph data at scale; schema evolution under load).
- Hands-on with LLM-based systems in production: prompt and model lifecycle, evaluation, cost control, and failure modes. You have opinions about when a model should and should not be in the loop.
- You have used AI coding agents seriously — not as autocomplete, but delegating whole work items — and you understand where they fail: context rot, confident wrong assumptions, rebuilding instead of reading, silent regressions.
- Strong writing. Your design docs are the primary interface between you and a fleet that reads them literally.
- Fluent English. Spanish is a plus.
Strongly preferred
- E-commerce platform integration experience (Shopify, WooCommerce, BigCommerce APIs, webhooks, metafields, app review).
- Cloudflare Workers / D1 or equivalent edge-and-SQLite architectures; AWS Bedrock; Linux operations on bare-metal servers.
- Experience with CI/CD design, branch protection, and merge-gate policy in high-throughput repos.
- SEO / AEO / structured-data (schema.org) familiarity.
- Prior startup experience at the stage where the architect also writes code.
How you work
- You resolve ambiguity with a written decision, not a meeting.
- You prefer evidence over narrative: when something is broken you find the failing unit, not the story about it.
- You are comfortable being the person who stops work that is out of scope, even when the proposal is clever.
- You can execute expansively once direction is set, and you will push back directly when you think the direction is wrong.
What we offer
- A rare seat: architect of both a product and the agent-native engineering system that builds it, at a company where that system is a strategic asset with commercial potential beyond PageMind's own product.
- Direct line to the CEO; real ownership of technical decisions.
- Competitive salary and equity.
- Remote-friendly within EU time zones; Flexible arrangement.
- Startup-program access across AWS, Azure, Google Cloud, and NVIDIA Inception; a budget for the models and hardware you need.
Process
- 30-minute call with the CEO.
- Take-home or live architecture exercise based on a real PageMind problem (e.g., "long-running daemons never reload merged code — design the mechanism that binds running processes to merged shas").
- Deep-dive with the platform team and an external technical advisor.
- References and offer.
Standard NDA, IP-assignment, and confidentiality terms apply to everyone with repository and customer-data access.
- PageMind SL — Barcelona. Send your CV, a link to something you designed and are proud of, and a short note on the worst failure mode you've seen in an AI-assisted engineering workflow to jaume.portell@pagemind.ai