Overview

This role involves designing and building multi-agent systems that solve complex business problems using foundation models and modern AI frameworks. You will focus on creating reliable, production-ready applications with robust evaluation and observability tools.

Summarised by InternDoor from EXL’s posting, in our own words. The posting itself is the source of truth, so check it before you apply.

Responsibilities

Build multi-agent systems that solve real business problems with explicit orchestration and state management
Work with foundation models and frameworks like LangGraph, LangChain, and Google ADK
Design evaluation tools and observability for production-grade AI applications

Requirements

Tick what you already have. It stays on this device and is only for you.

0 of 9 required ticked

Required

Experience Approximately 2-6 years of software, ML, AI, or product engineering experience, including meaningful hands-on work with Generative AI or LLM-powered applications
Programming Skills Strong Python programming skills and good software engineering fundamentals, including APIs, asynchronous programming, testing, Git, packaging, and debugging
LLM Concepts Solid understanding of LLM concepts including tokens and context windows, system and user prompts, structured outputs, tool/function calling, embeddings, retrieval, model selection, temperature and sampling, latency, and inference cost
Agent Orchestration Experience implementing explicit agent orchestration rather than only prompt chains or chatbots, including state, routing, delegation, tool permissions, retries, checkpoints, and failure recovery
RAG and Grounding Experience with RAG and grounding patterns using enterprise documents, structured data, vector search, semantic search, or knowledge stores
Integration Experience Experience integrating LLM applications with external APIs, business systems, databases, and developer tools; familiarity with MCP is valuable
AI Evaluation Understanding of AI evaluation techniques, including task-based evals, LLM-as-judge where appropriate, deterministic checks, test datasets, tracing, and regression testing
Production Concerns Understanding of production concerns including prompt injection, data leakage, access control, secrets, rate limits, token budgets, latency, reliability, observability, and cost management
Debugging Ability Ability to investigate ambiguous model or agent failures systematically and improve behaviour through changes to prompts, context, tools, orchestration, models, or application logic

Nice to have

Agentic Systems Built a multi-step or multi-agent system in which multiple specialised agents coordinate through an orchestrator, supervisor, graph, workflow, or shared state
Tool-Using Agents Experience designing tool-using agents that take actions rather than only generate text, with controls around what each agent can access and execute
Development Environments Familiarity with modern agent development environments and AI coding tools such as Google ADK, OpenAI Codex, Claude Code, and related SDKs or command-line development workflows
Durable Execution Experience with durable execution concepts such as checkpoints, resumability, idempotency, timeouts, retries, error routing, long-running tasks, and human-in-the-loop review
Agentic System Architecture Ability to explain the architecture of an agentic system end-to-end, including model choice, orchestration, state, memory, tools, identity, permissions, evaluation, monitoring, deployment, and operational support
Databricks/Snowflake Experience building agentic solutions within Databricks or Snowflake, including governed access to enterprise data, models, functions, or tools
Evaluation and Observability Exposure to Databricks Unity Catalog, Model Serving, Vector Search, AI Gateway / Unity Gateway, or Snowflake Cortex Agents, Cortex Analyst, Cortex Search, and Cortex AI Gateway
Evaluation Tooling Experience with MLflow, LangSmith, Langfuse, Arize Phoenix, OpenTelemetry, or similar evaluation and observability tooling

Read from EXL’s posting. Its own wording is the one that counts.

How to apply

Check what they ask for. Read the requirements above: EXL asks for 2–6 years.
Tailor your resume to the role. Put the work that matches these skills first. InternDoor’s resume tailor does it against this posting.
Apply on EXL’s careers site. The application is on forms.cloud.microsoft, not on InternDoor. We never ask for a fee.

Apply soon: postings like this can close without notice, and this one states no deadline.

Application details

Posted: 6 Nov 2026, on LinkedIn
First seen by InternDoor: 6 Nov 2026, within 11 minutes of being posted
Last checked: 6 Nov 2026: the application page loaded. That shows the page is up, not that EXL is still filling the role.
Application deadline: EXL has not stated one
Leaves InternDoor: By 6 Dec 2026, our 30-day limit for LinkedIn listings, or sooner if the application closes
Application page: forms.cloud.microsoft

What we checked, and what the dates mean

The date this listing leaves InternDoor is our own rule, not a deadline set by EXL. Postings can close earlier, at any time.

What “last checked” means: our checker opened the application page and it answered. That shows the page is up; it does not prove the role is still being filled, and EXL has not confirmed anything to us.

InternDoor has recorded 68 engineering postings from EXL in India since Sept 2026.

When this listing comes down, this address says the role has closed and lists open roles like it, instead of a dead end.

Hiring at EXL? Post a role on InternDoor, free

Similar roles

Full-time roles at other employers, closest first: the experience they ask for, the specialisation, the city, then the work mode. Internships are left out.