GenAI Agent Developer

Roche · Hyderabad

  • Experience5–8 yrs
  • SalaryNot disclosed
  • Work modeonsite
  • Levelsenior
  • Posted2 Sept 2026
Apply for this job

About Roche

Roche is hiring in Hyderabad in pharma biotech. This role looks for around 5+ years of experience.

Skills

  • prompt engineering
  • fine-tuning
  • retrieval-augmented generation
  • GraphRAG
  • schema-constrained outputs
  • function/tool-calling
  • semantic caching
  • LangChain
  • LangGraph
  • LlamaIndex
  • CrewAI
  • AutoGen
  • DSPy
  • FAISS
  • Milvus
  • Qdrant
  • Pinecone
  • Weaviate
  • Pgvector
  • vLLM
  • Hugging Face TGI
  • Ollama
  • NeMo Guardrails
  • Guardrails AI
  • Pydantic
  • Instructor
  • Ragas
  • DeepEval
  • Langfuse
  • LangSmith
  • Phoenix
  • Unstructured
  • Apache Tika
  • LlamaParse
  • PDFPlumber
  • Python
  • FastAPI
  • asyncio
  • REST APIs
  • GraphQL APIs
  • Git
  • CI/CD
  • Docker
  • Kubernetes
  • OpenTelemetry
  • Model Context Protocol
  • Agent2Agent communication standards
  • Amazon Bedrock
  • Amazon SageMaker
  • Amazon OpenSearch Service
  • AWS Step Functions

The role

A generative AI engineer at a pharmaceutical biotechnology company designs agentic workflows using retrieval-augmented generation, LangChain, and Amazon Bedrock, and builds evaluation and guardrail systems with Python and Kubernetes. The role develops autonomous LLM applications, manages vector search and model serving, and applies prompt engineering and cloud deployment practices.

Full job description

Responsibilities:

Agentic Architecture: Design, deploy, and scale multi-agent orchestration systems and autonomous workflows using cutting-edge frameworks.

Advanced RAG Pipelines: Build and optimize advanced retrieval-augmented generation (RAG) pipelines over massive, heterogeneous datasets (structured and unstructured).

State and Memory Management: Implement robust state management, short/long-term memory systems, and self-correction/reflection loops within agent networks.

Evaluation and Guardrails: Create and implement robust evaluation metrics, observability pipelines, and guardrails for content quality, hallucination reduction, bias mitigation, and safety standards.

Performance Optimization: Monitor and optimize AI inference cost, latency, throughput, token usage, and overall system reliability.

Security and Access Control: Implement robust access controls, data encryption, user authentication, and prompt injection mitigation across all LLM workflows.

Collaboration and Best Practices: Document and share reusable agent patterns, prompt libraries, and engineering components across cross-functional technical teams.

Requirements:

Demonstrated experience taking ownership of ambiguous tasks, successfully driving small to medium initiatives, and acting as a technical mentor.

Proven track record of engaging in knowledge-sharing initiatives, speaking at internal technical events, and navigating group dynamics in diversified settings.

Technical Skills:

GenAI Development: Advanced prompt engineering, fine-tuning, RAG/GraphRAG, schema-constrained outputs, function/tool-calling, and semantic caching.

Agentic Frameworks: LangChain, LangGraph, LlamaIndex, CrewAI, AutoGen, and DSPy.

Vector Databases: FAISS, Milvus, Qdrant, Pinecone, Weaviate, and Pgvector.

LLM Serving and Infra: vLLM, Hugging Face TGI, and Ollama (for local development).

Guardrails and Validation: NeMo Guardrails, Guardrails AI, Pydantic, and Instructor.

LLM Ops and Observability: Ragas, DeepEval, Langfuse, LangSmith, and Phoenix.

Data Parsing: Unstructured, Apache Tika, LlamaParse, and PDFPlumber.

Programming and DevOps: Python (FastAPI, asyncio), REST/GraphQL APIs, Git, CI/CD pipelines, Docker, Kubernetes, and OpenTelemetry.

Emerging Protocols: Model Context Protocol (MCP) and Agent2Agent communication standards.

AWS Ecosystem (Baseline Experience):

Amazon Bedrock: Foundation model access, custom configurations, and managed agent workflows.

Amazon SageMaker: Fine-tuning, hosting, evaluating, and deploying open-source LLMs.

Amazon OpenSearch Service: Vector search, hybrid search, and enterprise retrieval infrastructure.

AWS Step Functions: Multi-step orchestration and state machine management for hybrid AI/traditional pipelines.