Forward Deployed AI Architect

Most enterprise AI initiatives stall between pilot and production. The demo works; the path to a system that runs a real business process, under real controls, with a real owner, is where they get stuck. The Forward Deployed AI Architect is the person we place inside the client to get them unstuck.

You will be embedded with client business owners and engineering leaders, designing AI-native, Enterprise AI and Applied AI solutions — assistants, RAG systems, agents, agentic workflows, autonomous components — and taking them from pilot to production. You will carry the architect's responsibilities (framing the problem, fitting the business scope exactly, making trade-offs explicit, deciding what the system must not do) while sitting at the client's table, in their vocabulary, on their timeline.

The role is modeled on the forward-deployed engineering pattern but operates at the architecture level. It rewards people who like ownership, who can work with ambiguity without a home team beside them, and who measure themselves by what reached production at the client, not by what was presented.

Req# 1090267686

Responsibilities

• Embed with a client account, working directly with business owners, product leaders, engineering teams, security, risk, and operations to take AI solutions from pilot to production
• Design end-to-end AI-native architectures: orchestration, retrieval, integration with systems of record, agent autonomy and guardrails, identity and authorization, evaluation, observability, and cost model
• Own three artifacts that make your impact measurable: a capability economics model — what each AI capability costs and what it earns, so investment decisions are made on evidence; trust-rung evidence for every agent — the documented proof of how much autonomy each agent has earned and under what controls; a knowledge-graph slice of the client domain — the entities, relationships, and constraints the solution depends on, captured so the architecture and the business share one model
• Turn contradictory or incomplete requirements into explicit business decisions with named owners
• Decide, per component and per risk, how much can be delegated to AI — including AI coding agents in the delivery process — and where a human must stay in the loop
• Define how each system is evaluated: how the client will know an assistant or agent fits the business scope, not just that it produces fluent output
• Build proofs of concept and reference implementations when a decision needs evidence or a team needs an example of "good"
• Feed what you learn at the client back into EPAM's AI architecture practice: patterns, reference architectures, and what generation gets wrong

Requirements

• You have taken AI-native systems to production at a client or customer — RAG, assistants, agents or agentic workflows with real users and real business processes behind them — in a solutions architecture, customer engineering, forward-deployed, or consulting role, with architect-level responsibility for the system around the model
• You are a proven solution architect from a consulting or client-facing background who has started building with AI on your own initiative in the last year or so, and wants to make AI-native design at the client's side the center of your work
• A track record of client- or customer-facing architecture: you have framed problems for people who are not technologists, presented options with costs and risks, and got decisions made
• Solid solution or software architecture experience with delivery accountability — systems you designed that shipped and ran
• Depth in integration, distributed systems, cloud, and data-flow architecture at enterprise scale
• Comfort operating with ambiguity, conflicting stakeholders, and shifting requirements, without waiting for clarity before starting
• A habit of writing things down so decisions survive you: ADRs, design documents, evidence for autonomy and risk decisions
• Recent hands-on coding ability, enough to prototype and keep your constraints honest
• Willingness to be on-site with clients and to travel as the engagement requires

Nice to have

• Regulated-industry experience — financial services, healthcare and life sciences, insurance — including working with security, risk, and compliance functions
• Agent frameworks, tool and function calling, MCP, orchestration patterns, LLM observability
• Identity and authorization for agentic systems (OAuth 2.1, SPIFFE, AuthZEN, DPoP)
• Experience with cost and token economics, or with building business cases for AI capabilities
• Presales, proposal, or account-shaping experience
• Forward-deployed or field engineering background at an AI or data platform vendor