AI solution/ engineer

EPAM; United States; Hybrid; USD 150,000 - 190,000; Full time

Job summary

EPAM приглашает опытного специалиста по разработке и внедрению инфраструктуры оценки и тестирования для устойчивой продукции ИИ. Ваша миссия — создать платформу оценки, обеспечить доверие к градации и калибровать метрики против ручных меток, контролируя неопределенность поведения моделей.

Вы будете руководить командой, формируя стратегию оценки и регламент CI/CD. Требуются углубленные знания Python, опыт работы с LLM-диспетчерством, владение статистикой и методиками верификации.

Qualifications

• 5+ лет опыта в построении инфраструктуры оценки/тестирования, на выделенных гарнитурах и фреймворках.
• Опыт руководства командой и проектов.
• Экспертиза в AI-оценке, покрывающей качество вывода, поиск источников и регрессионный детектинг.
• Опыт в методологии LLM-диспетчера: дизайн рубрики, калибровка против ручных меток и осознание недостатков метода.
• Профильный уровень Python, глубокий для библиотек, знание TypeScript или Java
• Знания статистики для недетерминированных систем, включая выборку, доверительные интервалы, согласование между тестами и значимость.
• Навыки дизайна CI/CD-процессов для API, веб и дата-сайсов.
• Опыт работы с золотыми наборами и версионированием датасетов.
• Уровень английского не ниже Upper-Intermediate (B2).
• Будут плюсом: аудирование арабского языка и связанные методики оценки.

Responsibilities

• Set evaluation strategy: what gets measured, at which layer, with what methodology, and how results feed product decisions
• Build shared platform with evaluation harnesses, dataset versioning, and reporting читаемость
• Make grading trustworthy via judge model selection, rubric design, calibration
• Evaluate retrieval and agents for grounding/citation correctness, and multi-step reasoning
• Run online evaluation, drift detection, red-teaming and data-leakage tests
• Wire evaluation and tests as CI/CD gates
• Develop Arabic golden sets and judge calibration for Arabic
• Oversee online evaluation across AI services

Skills

Python

TypeScript

CI/CD

Statistics

LLM-evaluator

Rubric design

Calibration

Adversarial testing

Arabic evaluation

Security testing

Voice & conversational eval

Tools

LangSmith

Langfuse

Promptfoo

DeepEval

Braintrust

Job description

Задачи:

• Set evaluation strategy: what gets measured, at which layer, with what methodology, and how results feed product decisions
• Build the shared platform, including evaluation harnesses, golden-set management, dataset versioning, automated grading, regression detection, and reporting leadership can read
• Make grading trustworthy through judge model selection, rubric design, calibration against human labels, and knowing when automated grading cannot be trusted
• Evaluate retrieval and agents for grounding and citation correctness, tool-use validation, multi-step reasoning and failure recovery
• Develop Arabic golden sets and judges calibrated for Arabic rather than assumed to transfer from English
• Run online evaluation and drift detection
• Conduct prompt injection, jailbreak and data-leakage red-teaming
• Wire evaluation and test automation together as quality gates in CI/CD

Требования:

5+ years of experience building evaluation or test infrastructure others depend on, including harnesses, shared libraries and frameworks At least 1 year of relevant leadership experience Expertise in AI evaluation covering output quality, retrieval and grounding, and regression detection on non-deterministic behavior Proficiency in LLM-as-judge methodology: rubric design, calibration against human labels, and understanding its failure modes Advanced proficiency in Python, at the depth needed to build shared libraries, with intermediate competency in TypeScript or Java Knowledge of statistics for non-deterministic systems, including sampling, confidence intervals, inter-rater agreement and significance Skills in CI/CD framework design across API, web and data surfaces, with test selection, parallelization and flake management Background in golden sets and dataset versioning English proficiency at an Upper-Intermediate level (B2) or higher Будет плюсом: Familiarity with ragas, DeepEval, Promptfoo, Braintrust, LangSmith or Langfuse; Expertise in Arabic evaluation, including golden sets, dialect coverage, right-to-left validation and judge calibration; Background in adversarial and security testing, including red-teaming; Capability to perform voice and conversational evaluation, and performance testing against AI services; Experience delivering in a regulated or government environment

Условия:

Enjoyable working environment, whether it is the vibrant office or the comfort of your own home Opportunity to work abroad for up to two months per year Relocation opportunities within our offices in 55+ countries Corporate and social events Leadership development, career advising, soft skills and well-being programs Certifications, including GCP, Azure and AWS Unlimited access to EPAM's internal learning database Free English classes with certified teachers Discounts in local language schools, including offline courses for the Uzbek language Monetary bonuses for engaging in the referral program Medical & family care package Four trust days per year (sick leave without a medical certificate) Discounts for fitness clubs, dance schools and sports programs Benefits package (sports activities, a variety of stores and services)