Gen AI Architect

• Experience: 12-15years
• Location: Noida/Bangalore
• Requirements: Experience in Observability and OpenTelemetry - LLM/Applications
• Send resumes to: [Confidential Information] with below details:
• Name:
• Exp:
• CTC:
• ECTC:
• Notice period:
• Current location:

Job Description:

Level: 12+ years in software or platform engineering, including 3+ years designing

GenAI systems that other engineers shipped.

Own the architecture for LLM applications and the platform under them: models,

data, agents, evaluations, delivery, and the controls around them.

You will

Set the architecture for model access, retrieval, agent runtime, shared tools,

and one place to see traces and scores.

Choose when one agent is enough and when work splits across agents. MCP

is the tool boundary. Multi-agent handoff, including A2A, is the agent

boundary.

Define RAG so index lifecycle, citations, and freshness are part of the design,

and judge whether the enterprise data can support it.

Set standards others implement: OpenTelemetry GenAI as the span contract,

an evaluation approach so scores stay comparable, and LLMOps so prompts,

agents, and indexes are versioned, regression-tested, and rollback-able.

Place guardrails, tenancy, secrets, and retention in the platform. Decide what

an agent may read or change, and where a person must approve.

Shape inference for latency and cost: model routing, smaller models where

they are enough, caching, and batching where they help.

Write the architecture down and defend the tradeoffs with engineering and

with the people who own the business outcome.

Skills

A system you can walk through from request to stored trace, score, and cost,

on a cloud you have operated: AWS, Azure, or GCP, including Bedrock, Azure

OpenAI, Vertex AI, or Databricks.

One agent framework at design depth: LangGraph, CrewAI, AutoGen,

Semantic Kernel, OpenAI Agents SDK, or Google ADK. Familiarity with the

others is enough.

RAG architecture: hybrid retrieval, reranking, grounding, and the data work

underneath the index.

MCP and multi-agent design, including memory, tool scope, and human

approval. Direct A2A experience is a plus.

Evaluation and observability standards: golden sets, live signals, LLM-as-

judge, and GenAI spans for model, retrieval, tool, and agent steps. Operating

experience with MLflow, Arize Phoenix, LangSmith, or Langfuse is the

evidence. You will set the gen_ai.* contract here. You do not need to have

authored that spec elsewhere.

LLMOps: CI/CD for prompts and agents, regression gates, model and index

rollout, and rollback.

Platform controls: isolation between tenants and agents, guardrails, audit,

retention, and token cost.

Safety and reliability: prompt injection, tool abuse, timeouts, and a defined

path when the model is wrong.