Sr AI Architect/Engineer

Transaction Network Services

1d•$150,000 - $250,000

About The Position

Researches, designs, plans, develops and/or evaluates new, advanced network technologies by translating customer requirements or strategic business objectives into final product design. Work typically includes in-depth assessment of the customer’s networked application needs, providing network modeling and simulation testing, consulting to further clarify needs, and engineering high-speed technology network transports to deliver application functionality. TNS is moving from AI governance and foundations into execution. The platform building blocks are in place or underway — Microsoft 365, Copilot, Entra ID, SharePoint, and Anthropic's Claude models via AWS Bedrock. What we don't yet have is a dedicated person who can both build and operate the secure, production-grade AI platform that lets AI work safely against our data and embed with our lines of business to turn that capability into measurable value. This role does both. You'll be the technical engine and the on-the-ground partner who makes AI a productive, trusted tool across the enterprise — and who multiplies that impact by enabling others to build, not just building yourself.

Requirements

Strong software / solutions engineering background with hands-on experience building and deploying production AI/LLM applications — RAG, agents, tool use, orchestration, and evaluation.
Direct experience with a managed model platform in production; AWS Bedrock strongly preferred, along with the surrounding AWS estate (IAM, VPC, KMS, CloudWatch, PrivateLink).
Demonstrated ability to build and operate data connectivity to enterprise systems with permission-aware retrieval.
Working command of enterprise security and data governance: identity and access, data protection, encryption and key management, and AI-specific threat models.
Production operations instincts — you instrument what you build, and you've owned the monitoring and on-call reality of a live system.
Cost discipline: you treat inference spend as an engineering constraint and can show where the money goes.
Genuine business fluency and communication skills — you're comfortable sitting with finance, operations, or sales, understanding their problems, and translating them into solutions.
A demonstrated bias toward shipping and adoption, not just building demos.

Nice To Haves

Depth in the Microsoft / Azure ecosystem alongside AWS (Azure AI, Copilot Studio, Power Platform, Entra ID, SharePoint / Graph) — our environment spans both.
Experience in a regulated or security-sensitive industry — payments, telecom, or financial data.
Self-hosted or fine-tuned model experience, and a clear view of when it beats a managed API.
Infrastructure-as-code and CI/CD for AI systems (Terraform, pipelines, automated evals in the release path).
Change-management or internal-consulting experience.
Prior forward-deployed or customer-facing engineering experience.

Responsibilities

Build and operate the AI platform: Model platform and private hosting. Own our Claude deployment on AWS Bedrock end to end — model access and versioning, region and data-residency decisions, provisioned throughput vs. on-demand, and the private, VPC-isolated hosting posture that keeps our data inside our boundary and out of third-party training. Evaluate and manage additional models where they fit, and maintain a routing layer so applications aren't hard-wired to a single model or provider.
Harness and orchestration management. Build and maintain the application harness that sits between our users and the models: prompt and context management, tool and function calling, agent orchestration, retrieval pipelines, guardrails and content filtering, caching, and fallback behavior. Own the evaluation harness too — regression suites, quality benchmarks, and the release process that lets us upgrade models without breaking what's in production.
Data connectivity. Design and build secure, governed connections between AI systems and TNS data — Microsoft 365 and Graph, SharePoint, databases, data warehouses, APIs, and line-of-business systems. Own the retrieval architecture, indexing and freshness strategy, and the permission-trimming that ensures every user sees only what they're entitled to see, with no leakage across boundaries.
Security and access control. Architect least-privilege access across identity, network, and data layers — IAM and Entra ID, private networking, encryption and key management, secrets handling, and tenant isolation. Defend against AI-specific threats: prompt injection, data exfiltration through model outputs, insecure tool use, and over-permissioned agents. Partner with security and governance so every deployment is auditable and compliant by design, with a full audit trail of prompts, outputs, and data access.
Monitoring and observability. Instrument the platform for production operations — latency, error rates, throughput and quota headroom, output quality and drift, guardrail trips, and anomalous usage. Build the dashboards and alerting that let us find problems before users do, and the usage telemetry that tells us which solutions are actually delivering value.
Cost management. Own AI spend as a first-class engineering concern: token-level cost attribution by application and line of business, budgets and alerting, model right-sizing (using the cheapest model that meets the quality bar), prompt caching and context optimization, batch processing where latency allows, and provisioned-throughput decisions. Produce the chargeback and unit-economics reporting leadership needs to fund the portfolio with confidence.
Reusable foundations. Turn all of the above into patterns, templates, and guardrails so each new use case is faster, cheaper, and safer to build than the last.
Ship value with the business (forward-deployed engineering): Embed directly with teams across our lines of business (Communications, Waypoint, Payments, Corporate/Shared) to identify high-value AI and automation opportunities.
Prototype quickly, validate with real users, and take the strongest ideas to production.
Drive adoption — measure whether solutions are actually used and delivering value, and iterate until they are.
Enable others: seed and support a network of "AI champions" in each line of business, equipping non-specialists to build their own automations on your patterns and under appropriate oversight.