AI Systems Engineer

Vinmar International

2d•Remote

About The Position

Vailent is building the AI infrastructure for the materials industry (chemicals, polymers, elastomers, rubber), replacing fragmented tools and flat files with a unified system that transforms interactions, transactions, and physical assets into usable commercial data. The materials industry is foundational to the physical economy but remains under-instrumented. Vailent aims to transform this industry by capturing multi-modal industry context across software and hardware. This role is for a full-stack platform engineer responsible for operating a multi-application B2B platform end-to-end. This includes directing AI agents, verifying operations in the real environment, and owning the entire stack: cloud infrastructure, backend, frontend, data, and deep enterprise-ERP integration. The focus is on operating AI, not just writing code with it, involving decomposition, distribution, adversarial verification, and shipping. The position requires managing a B2B platform with approximately ten applications on a shared cloud backbone, including deep integration with customer enterprise systems like SAP/ERP. The engineer will own the full stack from Terraform and IAM to React components and SAP RFC calls. The key differentiator is the ability to comprehend the entire platform and orchestrate AI agents effectively while maintaining architectural coherence across multiple repositories. AI orchestration is central to the role's scope and success, requiring fluency and discipline for safe operation.

Requirements

Fluent AI orchestration: ability to run agents in parallel, isolate work in worktrees, and verify output adversarially.
Genuine full-stack and infrastructure range: comfortable working across React components, Postgres RLS policies, and Terraform modules within the same day.
Systems debugging instinct: ability to trace root causes across service boundaries (auth, pagination, dependency conflicts, integration mismatches).
Evidence reflex: distrust of green badges, demand for real fixtures, and proof of concepts via working screenshots, read-back records, or live payloads.
Self-correction: ability to describe a time of reversing a confident conclusion based on evidence.
Automation reflex: instinct to turn recurring work into reusable Claude skills, hooks, and commands.
Operating discipline: includes working read-only until authorized, copying proven patterns, enhancing in place, using precise language, and ensuring clean handoffs.
Thick skin and plain speech: ability to receive blunt, fast feedback well and explain reasoning simply.

Nice To Haves

Enterprise ERP/SAP depth, including RFC/BAPI, ABAP, customer & order master data, or the ability to reverse-engineer a customer's system to that depth.
Experience building scaffolding that enhances agent and engineer effectiveness (e.g., nav maps, indexes, skills, guard hooks, templates).
Architectural taste under constraint: ability to choose boundaries that maintain flat future costs and articulate why rewrites or scatter solutions are suboptimal.
Multi-tenant/B2B context: understanding of tenancy isolation, per-tenant configuration, and associated failure modes.
Compliance fluency: comfort with GDPR, SOC 2, ISO 27001, ROPA, control mappings, and runbooks.
Design-system literacy: preference for tokens over hardcoded values, and ability to perform UX and UI passes on own work.

Responsibilities

Own the platform end to end, including multiple applications and shared SDKs on a single cloud backbone (React/TypeScript front ends, FastAPI/Python services, Terraform/IAM/ECS infrastructure, and a shared design system).
Set up infrastructure and environments from scratch, including new services, cloud accounts, tenants, connectors, data syncs, and migrations (including cross-region), ensuring they are provisioned and proven.
Direct fleets of coding agents to decompose cross-repo changes into disjoint tasks, fan them out to parallel agents in isolated worktrees, run adversarial multi-reviewer passes, and reconcile results.
Integrate deeply with enterprise systems like SAP/ERP via RFC/BAPI, including reading and authoring ABAP, reverse-engineering business rules, handling sales-order and customer-master flows, currency/unit/sales-area mapping, and idempotent event sync.
Architect multi-tenant data using PostgreSQL row-level security, JSONB-backed tenant-extensible capability platforms, careful migrations, and a graph database where appropriate.
Ship at volume across multiple repositories in a working session while maintaining architectural coherence, ensuring CI is green, and deployments are verified.
Author documentation including specs, design docs, discovery-question sets, and runbooks to facilitate understanding and resumption of work by others.
Build tooling to enhance AI effectiveness, such as per-codebase navigation maps, documentation indexes, guard hooks, and custom skills.
Automate recurring tasks by capturing them as reusable Claude skills, hooks, or slash commands.
Review code adversarially to catch regressions, differentiate rendering from data correctness, refute false blockers, and manage shared state carefully.
Prove work in the real environment, ensuring that 'done' means demonstrated, not asserted, and that data correctness is paramount.
Diagnose issues thoroughly before making changes, especially in live environments, by identifying root cause and proposing a fix before proceeding.
Copy and adapt proven patterns from existing working examples rather than inventing new approaches.
Enhance existing systems in place by generalizing paths and adding optional parameters, rather than creating parallel reimplementations.
Identify and flag non-scalable aspects as explicit, revisit-able debt.
Verify findings adversarially and be willing to reverse conclusions based on evidence.
Use precise language and correct loose terminology.
Leave a clear trail of documentation, including specs, runbooks, tracked tickets, and durable notes, for future reference and handoffs.