Lead Full-Stack, Data and AI Agent Engineer

EPAM·Argentina, Colombia, Mexico, Brazil, Chile·Удалённо·сегодня

We are looking for a Lead Full-Stack, Data and AI Agent Engineer to build a production-grade agentic AI platform for upstream oil & gas operations (ESP, gas lift, plunger lift, chemical injection). As a Lead Full-Stack, Data and AI Agent Engineer, you will evolve the deep-agent core that converts petroleum engineers’ natural-language questions into safe, schema-aware SQL and multi-step reasoning workflows, and help the team ship reliably.

Responsibilities

Lead ownership of the deepagents/LangGraph agent runtime, extending agent tools, subagents, and safety middleware
Design FastAPI endpoints using async I/O (asyncpg/psycopg3) and SSE streaming to deliver real-time agent responses
Implement SELECT-only, injection-safe SQL generation across per-tenant PostgreSQL schemas
Integrate AWS Bedrock (Claude via langchain-aws) and manage LLM factory/model routing decisions
Instrument agent runs with MLflow and contribute improvements to the eval framework
Collaborate on RAG retrieval using pgvector, FAISS, and S3-backed document storage

Requirements

Proven track record with 5+ years of backend engineering in Python, including strong async/await skills and FastAPI or a comparable framework
Hands-on experience with AWS core services such as Bedrock, RDS, and S3, following a strict no-hardcoded-credentials approach
Practical experience using LangGraph, LangChain, or a similar agent-orchestration framework
Solid SQL/PostgreSQL experience, with confidence owning safety-critical code such as SQL injection prevention and tenant isolation
Deep understanding of LLM APIs (Bedrock, OpenAI, Anthropic) and prompt engineering techniques
Strong CI/CD and testing discipline, including lint/format gates, pytest/Jest, and GitHub Actions
Security-first mindset for tenant isolation, including tenant isolation, SELECT-only SQL, and no cross-tenant leakage
Observability and eval mindset using MLflow or equivalent, measuring quality, latency, and tokens before shipping
Ability to learn oil & gas / production-ops terminology (ESP, gas lift, decline curves) even without prior background
English proficiency for client-facing communication; Spanish is a nice-to-have

Nice to have

Working knowledge of TypeScript/React/Next.js, including SSE event formats, content types, and payload shapes for cross-stack debugging
Familiarity with uv, Ruff, and pytest workflows
Experience with MLflow
Familiarity with Redis/APScheduler
Background in oil & gas domain knowledge