[Hiring] Applied AI Engineer @upstream

Role Description

You are our first dedicated Applied AI Engineer, and your remit is the whole business: sales, customer success, marketing, delivery, finance and operations. Anywhere someone is doing work a well-designed agent or workflow should be doing, that's yours.

Critically, you are the architect. Work will reach you as a problem, not a specification:

• "renewal risk is invisible until it's too late"
• "delivery handovers lose information every time"
• "this report takes two days a month to produce"

From there the system is yours:

• Interrogate the brief
• Work out what it should actually do
• Choose the approach and the tooling
• Design how it sits across the platforms we already run
• Build it, launch it and own it in production

Nobody will hand you a solution design, and nobody will review your architecture for you. That is the job, and it's why we're hiring at this level rather than for a pair of hands.

It also means you'll work directly with the people who own the results your systems affect — short feedback loops, no committee, and a very clear read on whether what you built actually mattered.

Wide surface, real autonomy, and everything you ship gets used the next morning.

What you'll own

• Production AI agents and routines — taking prototypes to monitored, reliable, production-grade systems, starting with the commercial suite already in flight
• A reusable skills and prompt library for drafting, analysis, reporting and document generation, shared across functions
• The internal integration layer across monday.com, HubSpot, Aircall, Microsoft 365 and Guidde
• Commercial systems — pipeline and portfolio reporting, account health signals, onboarding flows, executive business review generation
• Marketing and go-to-market systems — content pipelines and campaign operations
• Delivery and operations systems — project reporting, resourcing visibility, document and scope production, handover automation
• Finance and administrative workflows — reconciliation, reporting and the recurring manual work that currently eats hours
• Documentation, standards and internal enablement — everything you build is documented, supportable and teachable to the people who use it

Qualifications

• Demonstrable fluency with modern LLM tooling — agent design, prompt and context engineering, MCP, evaluation and iteration. Claude and Claude Code experience is a strong advantage
• Model-agnostic judgement — the ability to assess Claude, OpenAI, Gemini and whatever lands next on capability, cost, latency and fit, pick the right one for each job, and re-test that decision as the landscape shifts
• Strong automation and integration skills across SaaS APIs (REST, webhooks, authentication) and iPaaS platforms such as Make, Zapier or n8n
• The technical ability to ship and support production systems: Python/TypeScript, git, cron, CI/CD, Linux, error handling and logging
• Confidence working with data across systems — extracting, transforming and reconciling records from multiple platforms
• The ability to architect a solution from a vague problem statement — scoping it, choosing the approach, and being able to explain and defend the design to both technical and non-technical stakeholders
• A portfolio of things you have actually launched, that other people still use
• The judgement to push back on a brief, propose a better approach, and know when not to build something
• Comfort working across unfamiliar business domains — you'll be automating finance one month and delivery the next
• Written English strong enough to work asynchronously with a senior team across time zones
• The security-mindedness to work inside tenant boundaries with customer data

Nice to have

• monday.com or HubSpot platform experience (apps framework, GraphQL APIs)
• Prior work inside a consultancy or agency
• Light front-end for internal tools
• Experience with documentation or video documentation tooling

What this role isn't

• No billable client delivery
• No model training from scratch
• No general IT helpdesk
• Your entire remit is making one company measurably more effective

What success looks like

• First 30 days — access established, the systems we already run documented and understood, and the backlog assessed and sequenced with your recommendations on what ships first
• First 90 days — Existing AI based workflows optimised and performing in a production environment— monitored, logged and reliable without supervision — plus two new systems live and in daily use
• First 6 months — the same patterns extended beyond the commercial team into at least two other functions, with a documented, reusable foundation others can build on
• First 12 months — a maintained internal AI platform spanning the business, with measurable hours returned across functions and a clear roadmap you own