Staff Applied AI Engineer
About the Client
We are sourcing on behalf of Alpheya, a B2B WealthTech startup based in Abu Dhabi and backed by BNY Mellon and Lunate. The company has raised $300M to build a state of the art wealth technology platform. Their mission is to power and grow their clients’ Wealth franchises through differentiated experiences, financial solutions, and insights. Their digital wealth management platform will enable banks and other financial institutions in the Middle East to grow and further penetrate affluent, HNW and UHNW investor segments. While still leveraging the capabilities and knowledge of large organizations, this fintech is a startup with truly cross-functional and agile teams.
The Job
We have built an AI agent orchestration platform that already performs real engineering, devops, and security work across Slack, GitHub, Kubernetes, source code, operational data, and internal tools. A user can ask the agent to investigate an incident, inspect a deployment, query an internal system, explain a code path, implement a change, or test the platform. The request becomes an isolated execution session with access to approved tools and environments. The agent works through the configured workflow and reports the outcome back to the user.
We are hiring a Staff Applied AI Engineer to build that product and deploy it with our first enterprise customers. You will own substantial parts of the core platform, and spend time inside customer environments making the system useful in practice. This is a product engineering role with a forward-deployed component. It is not a pre-sales role, a prompt-writing role, a research position, or a managerial position. You will write production code and remain accountable for how it behaves in customer environments.
What You Will Own
Product Runtime
- The lifecycle of an agent request from Slack or another entry point through dispatch, execution, tool calls, and final reporting.
- Isolation and routing of work across organizations, environments, and execution planes.
- Durable session behavior, including retries, timeouts, cancellation, recovery, and human approval.
- The interfaces through which agents discover and call approved tools, including MCP servers and internal APIs.
- Identity, authorization, secrets, audit history, and policy enforcement around agent actions.
- Operational visibility into what an agent did, why it failed, and how to recover it.
- Packaging and upgrade paths for vendor-managed, customer-hosted, and hybrid deployments.
Customer Deployments
- Technical discovery with customer engineering, platform, security, and operations teams.
- Selecting the first one or two workflows that are valuable, bounded, and realistic to operate.
- Connecting the platform to the customer’s Slack, GitHub, Kubernetes clusters, data systems, and internal tools.
- Designing the access model for private networks and sensitive systems without routing unnecessary customer data through shared infrastructure.
- Taking each workflow from a successful prototype to a supported production capability.
- Feeding reusable requirements back into the core product instead of accumulating customer-specific forks.
Agent Quality
- Task-specific evaluations for correctness, completion, safety, and usefulness.
- Context and tool design for long-running engineering and operational tasks.
- Guardrails and approval points for actions with material impact.
- Investigation of failures across models, orchestration, tools, permissions, networks, and infrastructure.
- Clear evidence for whether a workflow is performing better than the process it replaces or assists.
Initial Outcomes
Within the first 6 to 12 months, we expect you to help deliver:
- A repeatable path for onboarding a new organization without a bespoke codebase.
- At least one production workflow operating against a customer’s private systems.
- A deployable execution environment with explicit identity, tool access, and network boundaries.
- End-to-end tracing and recovery for agent sessions and tool calls.
- A practical evaluation suite for the first customer workflows.
- A clear product boundary between shared platform capabilities and customer configuration.
- Product improvements derived from the first deployments that reduce the effort required for the next customer.
What We Need
You should have:
- A strong production software engineering background in backend, platform, infrastructure, or distributed systems.
- Direct experience building an LLM-powered product or agent that used tools and operated beyond a notebook or demo.
- Experience operating Kubernetes and cloud infrastructure, including debugging applications across networking, identity, permissions, and runtime boundaries.
- The ability to design APIs, asynchronous workflows, durable state, and failure recovery.
- Strong engineering judgment around security, reliability, and the blast radius of automated actions.
- The ability to work directly with customers, reduce ambiguous problems into concrete requirements, and then implement the solution yourself.
- The discipline to generalize what should become product while keeping customer-specific configuration appropriately isolated.
- Clear written and verbal communication with both engineers and decision-makers.
We do not require a specific number of years or a particular degree. We do require evidence that you have independently shipped and operated meaningful production systems. Please, be ready to discuss them during your interview.
Strong Signals
- Forward-deployed engineering or another role where you personally built software inside complex customer environments.
- Developer tools, coding agents, CI/CD systems, source-control integrations, or infrastructure automation.
- Multi-tenant SaaS, customer-hosted enterprise software, or control-plane and execution-plane architectures.
- Kubernetes controllers, multi-cluster platforms, workflow engines, or secure remote execution.
- MCP or another structured mechanism for exposing tools to AI agents.
- Enterprise identity, private networking, secrets management, authorization, and audit systems.
- Deployments in financial services or another regulated industry.
- Customer work across the Middle East, UK, Europe, or United States.
This Role Is Probably Not a Fit If
- Your AI experience is limited to prompt experimentation or model training without production software ownership.
- You want to hand customer requirements to a separate engineering team rather than implement them yourself.
- You are uncomfortable debugging infrastructure, permissions, networking, and integrations alongside application code.
- You prefer a narrowly defined backlog over owning an ambiguous problem from discovery through production.
- You optimize for a single customer implementation even when it creates a long-term product fork.
Technology
The current platform is built around Go, Kubernetes, Slack, GitHub, cloud infrastructure, and tool integrations including MCP. Customer deployments may also require Python, TypeScript, databases, observability systems, identity providers, and private networking. You do not need to arrive as an expert in every component. You do need to be comfortable crossing these boundaries and learning the parts required to deliver a complete system.
How We Work
- We start with a real workflow and a measurable outcome.
- Engineers own systems through deployment and production operation.
- Customer delivery and product development are part of the same feedback loop.
- We move quickly, but not by hiding reliability or security work.
- We communicate decisions, evidence, and trade-offs directly.
- We prefer a smaller number of production capabilities over a large collection of demos.
Privacy Notice
By applying to this position, you are granting us permission to process your CV and keep your profile on file for consideration for this and future opportunities.
