EPAM Systems
1 month ago
Lead AI Engineer, Agentic and RAG Systems
Sign up to save this job, get alerts, and apply with an optimized CV.
Company information
- Company
- EPAM Systems
- Location
- Georgia
- Posted
- 1 month ago
Job description
We are looking for a seasoned Lead AI Engineer who architects, builds, and operates production GenAI platforms – agentic workflows, RAG pipelines, and LLM-backed services with real users and real SLAs – while leading engineers and setting the technical direction across multiple workstreams.
This is an engineering leadership role, not a research role. The bar is reliability, latency, cost, observability, and safe deployment at scale, with end-to-end ownership from architecture through on-call, and accountability for the technical quality and delivery of the team. Typical workloads include enterprise knowledge platforms, conversational analytics, agentic automation, and LLM-augmented data products.
Responsibilities
- Own the end-to-end architecture of GenAI platforms across multiple services and teams, defining standards, patterns, and reference implementations
- Lead the design of agent orchestration (graph/state, conditional routing, tool calling, memory, checkpointing) in LangGraph / LangChain or equivalent, and set best practices for the team
- Architect production RAG end-to-end: chunking, embeddings, vector stores, hybrid retrieval, reranking, caching, and grounded synthesis – and mentor engineers in building it
- Drive the design and delivery of Python / FastAPI services – async, SSE streaming, session handling, and structured error contracts – establishing service templates and conventions
- Define the observability and evaluation strategy (MLflow, OpenTelemetry, or equivalent) for accuracy, cost, and regression across the platform
- Own the deployment platform on Docker + Kubernetes (EKS/AKS/GKE) with CI/CD, test, eval, and canary gates – setting release standards for AI systems
- Lead LLM cost engineering strategy – model routing, prompt optimization, caching, token accounting, and build-vs-buy decisions at portfolio level
- Establish GenAI safety & governance practices: hallucination control, prompt-injection defense, PII handling, and HITL where required
- Partner with data engineering leadership on semantic layers and pipelines (PySpark / SQL where applicable), and align roadmaps across teams
- Mentor and grow senior and mid-level engineers through design reviews, pairing, and technical coaching; conduct hiring and technical interviews
- Represent engineering in conversations with clients, product, and executive stakeholders; translate business goals into technical strategy and delivery plans
Requirements
- 6+ years in software engineering, with 3+ years shipping production LLM / agentic systems (not POCs or research)
- 1+ years of experience leading engineers or technical workstreams
- Proven track record of owning architecture for multi-service GenAI or distributed systems in production
- Expert-level proficiency in Python and FastAPI (async, REST, SSE)
- Deep production expertise in LangChain and LangGraph (or equivalent serious production experience with LlamaIndex, AutoGen, or MCP stacks)
- Strong background in production RAG: embeddings, chunking, and hybrid retrieval with reranking and caching – with the ability to define standards across teams
- Advanced skills in vector databases such as Pinecone, Weaviate, pgvector, OpenSearch, or Databricks Vector Search
- Hands-on production experience with at least one major LLM provider – AWS Bedrock (preferred), OpenAI / Azure OpenAI, or Anthropic – including model selection, routing trade-offs, and multi-provider strategy
- Strong competency in Kubernetes and Docker in real production environments (EKS/AKS/GKE), including platform-level decisions
- Deep expertise in cloud engineering on AWS, including cost, security, and scalability trade-offs
- Solid command of observability and tracing tools (MLflow, LangSmith, OpenTelemetry), evaluation harnesses, and latency/cost ownership at platform scale
- Experience designing and owning CI/CD for AI systems (GitHub Actions, Jenkins, or equivalent) with test/eval gates
- Demonstrated experience mentoring engineers, leading design reviews, and driving technical decisions across teams
- Strong written and spoken English (B2+ level); able to lead design discussions, present to senior stakeholders, and influence technical direction with clients and executives
Nice to have
- Databricks depth – MLflow (tracking & serving), Vector Search, Unity Catalog / Metric Views, PySpark / SQL
- Experience with LLM fine-tuning – PEFT, LoRA, QLoRA – and the ability to guide build-vs-fine-tune-vs-prompt decisions
- Strong understanding of MCP servers and tool integration patterns
- Expertise in GenAI governance & FinOps – auditability, prompt-injection hardening, PII, and token cost in regulated environments
- Background in classical ML / DL – NLP, BERT-family, time-series, and CV
Required skills
- docker
- databricks
- accuracy
- mentoring
- ci/cd
- security
- reliability
- governance
- sql
- python
- aws
- kubernetes
- nlp
- llm
- azure openai
- software engineering
- standards
- caching
- jenkins
- github actions
- eks
- research
- vector databases
- embeddings
- genai
- architecture
- best practices
- observability
- mcp
- tool calling
- data engineering
- pyspark
- fastapi
- product
- aks
- scalability
- rest
- pipelines
- cloud engineering
- distributed systems
- openai
- langgraph
- business goals
- clients
- slas
- design reviews
- llamaindex
- autogen
- hiring
- async
- finops
- mlflow
- cv
- langchain
- rag pipelines
- opensearch
- anthropic
- pocs
- regression
- senior engineers
- technical quality
- patterns
- gke
- engineering leadership
- technical interviews
- opentelemetry
- regulated environments
- chunking
- pinecone
- roadmaps
- aws bedrock
- executive stakeholders
- agent orchestration
- prompt optimization
- latency
- memory
- weaviate
- lead ai engineer
- agentic workflows
- unity catalog
- ai systems
- on-call
- lora
- technical strategy
- semantic layers
- cost engineering
- conventions
- delivery plans
- langsmith
- pgvector
- time-series
- pii handling
- written english
- llm fine-tuning
- dl
- mcp servers
- spoken english
- end-to-end ownership
- design discussions
- reranking
- auditability
- mentoring engineers
- peft
- agentic automation
- qlora
- pairing
- hitl
- service templates
- model selection
- classical ml
- platform scale
- technical coaching
- production platforms
- b2+ level
- vector stores
- conversational analytics
- evaluation harnesses
- safe deployment
- enterprise knowledge platforms
- llm-augmented data products
- reference implementations
- checkpointing
- production rag
- hybrid retrieval
- grounded synthesis
- sse streaming
- session handling
- structured error contracts
- deployment platform
- test gates
- eval gates
- canary gates
- release standards
- llm cost engineering
- model routing
- token accounting
- build-vs-buy decisions
- genai safety
- hallucination control
- prompt-injection defense
- mid-level engineers
- technical workstreams
- multi-service genai
- databricks vector search
- llm provider
- multi-provider strategy
- tracing tools
- latency ownership
- cost ownership
- driving technical decisions
- presenting to stakeholders
- influencing technical direction
- mlflow tracking
- mlflow serving
- metric views
- build-vs-fine-tune
- prompt decisions
- tool integration patterns
- genai governance
- prompt-injection hardening
- bert-family
Interested in this position?
Create your free account and tailor your CV to match this job.