Capgemini - Generative AI Developer - LLM

Job Description :

Hands-on GenAI Developer to build and optimize Generative AI applications and pipelines. This role focuses on execution (coding, integration, monitoring, and performance tuning of LLM-powered workflows) under the guidance of the Architect.

Key Responsibilities :

- Strong programming skills in Python for Agent development.

- Integrate LLMs (Gemini, Llama, OpenAI, Claude, Mistral) and embedding models into applications.

- Develop agentic workflows using LangChain, LangGraph, Google SDK and CrewAI applying patterns like Planner-Executor, Reflection Loops, and Multi-Agent Orchestration.

- Build data pipelines for ingestion, preprocessing, chunking, enrichment, embeddings, and vector indexing.

- Implement coarse-grained controls (e.g., data masking, compliance filters, security checks).

- Prototype in Jupyter Notebooks and transition solutions to production environments.

- Apply LLMOps practices: instrumentation, monitoring, evaluation automation, and feedback loops.

- Collaborate with Architects to integrate knowledge graphs where applicable.

- Deploy and maintain solutions on GCP using GPU/compute environments.

- Develop agentic flow using A2A (Agent2Agent) protocol

- Integrate with various tools using MCP (Model Context Protocol)

Required Skills & Experience :

- Must have worked on development of Agents using LangChain, LangGraph and Google ADK.

- Experienced in developing Agentic flows using A2A and integrating with tools using MCP.

- Strong Python skills; working knowledge of JavaScript.

- Experience with GCP and Jupyter Notebooks.

- Hands-on with LLMs, embeddings, vector databases, and agentic workflows.

- Familiarity with security guardrails and compliance in GenAI systems.

- Proven experience in LLMOps (logging, monitoring, evaluation pipelines).

- Knowledge of knowledge graphs is a plus.

- Strong debugging and problem-solving abilities.

Good-to have :

- Knowledge in observability using LangSmith and hosting agents on LangServe