Job description

Enterprises are past the point of asking whether AI belongs in their systems. The question now is which AI capabilities should exist, how much autonomy each one earns, how they integrate with systems of record and real business processes, and how anyone will know they are working. Answering those questions well - and writing the answers down so that engineering teams and coding agents can build against them - is the AI Architect's job. You will design the target architecture of AI-native applications, Enterprise AI and Applied AI solutions for large clients across industries, from assistants and RAG systems to agents, agentic workflows and autonomous components. You will decide what the system should do and, just as importantly, what it should not. You will define the evaluations, guardrails, and constraints that shape what gets built, whether the builder is an engineer or an AI coding agent. This is an architecture role. The output is decisions, structure, and the reasoning behind them - not code volume. You will still prototype when a decision needs evidence, but most of your time goes to framing problems, resolving conflicting requirements, making trade-offs explicit, and getting the right people to own them. Req# 1090267686 Responsibilities: • Design end-to-end architectures for AI-native solutions: orchestration, retrieval, tool and system integration, agent autonomy and guardrails, identity and authorization, evaluation, observability, and cost model • Fit each solution exactly to the business scope - no more, no less - and be able to explain what you deliberately left out and why • Turn incoherent or contradictory requirements into explicit business decisions: surface them, reframe them in business terms, price the options, and get a named owner to decide • Make and document architectural trade-offs (build vs buy, model selection, retrieval strategy, autonomy level, cost vs quality) in a form that stakeholders can challenge and future teams can trust • Define the evaluation strategy for each system: how we will know that an assistant or agent fits the business scope, not just that it produces fluent output • Author the constraints, guidelines, and architectural fitness functions that engineering teams and coding agents build within, and revise them as the organization learns what generation gets wrong • Decide where humans must stay in the loop - regulated decisions, data of record, security boundaries, anything with legal or contractual consequence • Work within EPAM delivery teams and in presales: shape proposals, present options and costs to client executives, and review designs across teams • Build proofs of concept and reference implementations when, and only when, a decision needs evidence or a team needs an example of "good" • Contribute to EPAM's AI architecture practice: reference architectures, reusable patterns, and mentoring architects who are transitioning into AI-native design Requirements: • You have architected AI-native systems in production - RAG, assistants, agents or agentic workflows with real users or real business processes behind them - and you held architect-level responsibility for the system around the model • You are a proven solution or software architect with a strong production track record who has started building with AI on your own initiative in the last year or so - a side project, a proof of concept, an internal tool, published writing - and wants to make AI-native design the center of your work • Solid experience as a solution, software, or enterprise architect with delivery accountability: systems you designed that shipped, ran, and were maintained over time • Depth in distributed systems, integration, cloud, and data-flow architecture at enterprise scale • Demonstrated ability to work with ambiguous, conflicting requirements and to make trade-offs explicit rather than resolving them silently • Experience presenting options, costs, and risks to non-technical decision makers and getting decisions made • A habit of writing things down: architecture decision records, design documents, reference architectures that other teams built from • Recent hands-on coding ability, enough to prototype and to keep your constraints honest • Understanding of how AI-assisted development changes the architect's role: what to specify, what to evaluate, and where to keep human judgment Nice to have: • Experience in regulated industries - financial services, healthcare and life sciences, insurance • Agent frameworks, tool and function calling, MCP, orchestration patterns, LLM observability • Identity and authorization for agentic systems (OAuth 2.1, SPIFFE, AuthZEN, DPoP) • Ownership of cost and token economics for a production AI system • Consulting or pre-sales experience • Published writing, talks, or open-source contributions on AI architecture

Requirements

Must-haves

Req# 1090267686
Design end-to-end architectures for AI-native solutions: orchestration, retrieval, tool and system integration, agent autonomy and guardrails, identity and authorization, evaluation, observability, and cost model
Fit each solution exactly to the business scope - no more, no less - and be able to explain what you deliberately left out and why
Turn incoherent or contradictory requirements into explicit business decisions: surface them, reframe them in business terms, price the options, and get a named owner to decide
Make and document architectural trade-offs (build vs buy, model selection, retrieval strategy, autonomy level, cost vs quality) in a form that stakeholders can challenge and future teams can trust
Define the evaluation strategy for each system: how we will know that an assistant or agent fits the business scope, not just that it produces fluent output
Author the constraints, guidelines, and architectural fitness functions that engineering teams and coding agents build within, and revise them as the organization learns what generation gets wrong
Decide where humans must stay in the loop - regulated decisions, data of record, security boundaries, anything with legal or contractual consequence
Work within EPAM delivery teams and in presales: shape proposals, present options and costs to client executives, and review designs across teams
Build proofs of concept and reference implementations when, and only when, a decision needs evidence or a team needs an example of "good"
Contribute to EPAM's AI architecture practice: reference architectures, reusable patterns, and mentoring architects who are transitioning into AI-native design
Solid experience as a solution, software, or enterprise architect with delivery accountability: systems you designed that shipped, ran, and were maintained over time
Depth in distributed systems, integration, cloud, and data-flow architecture at enterprise scale
Demonstrated ability to work with ambiguous, conflicting requirements and to make trade-offs explicit rather than resolving them silently
Experience presenting options, costs, and risks to non-technical decision makers and getting decisions made
A habit of writing things down: architecture decision records, design documents, reference architectures that other teams built from
Recent hands-on coding ability, enough to prototype and to keep your constraints honest
Understanding of how AI-assisted development changes the architect's role: what to specify, what to evaluate, and where to keep human judgment

Nice-to-haves

Experience in regulated industries - financial services, healthcare and life sciences, insurance
Agent frameworks, tool and function calling, MCP, orchestration patterns, LLM observability
Identity and authorization for agentic systems (OAuth 2.1, SPIFFE, AuthZEN, DPoP)
Ownership of cost and token economics for a production AI system
Consulting or pre-sales experience
Published writing, talks, or open-source contributions on AI architecture

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

aillm

Platform: epam
Effort: medium