AI Agentic engineer Full Time

Tata Elxsi is among the world’s leading providers of design and technology services across industries, including automotive, broadcast, communications, healthcare, and transportation. Tata Elxsi works with leading OEMs and suppliers in the automotive and transportation industries for R&D, design, and product engineering services from architecture to launch and beyond.

Role Summary

We are seeking a Senior AI Engineer who bridges two distinct but complementary disciplines: the rigorous, production-grade deployment of classical ML models and LLM-powered applications, and the rapidly evolving field of Agentic AI — designing, orchestrating, and operating systems where autonomous agents reason, plan, use tools, and collaborate to accomplish complex multi-step goals. You will have delivered real AI systems into production — not just prototypes — and understand the full engineering discipline that entails: reliability, observability, safety guardrails, latency management, cost control, and continuous evaluation. In this role you will architect Agentic AI solutions for enterprise clients, implement RAG pipelines and fine-tuned LLMs, design multi-agent workflows using frameworks such as LangGraph, AutoGen, or CrewAI, and embed AI capabilities into business-critical applications. You will also maintain and evolve a portfolio of traditional ML models — from training and evaluation through MLOps-automated deployment and monitoring in production.

Key Responsibilities

Agentic AI System Design & Engineering

• Architect and implement multi-agent AI systems using LangGraph, AutoGen, CrewAI, or Semantic Kernel — designing agent topologies (supervisor, hierarchical, collaborative peer), tool registries, memory architectures (short-term context, episodic memory, long-term vector stores), and inter-agent communication protocols for enterprise-grade reliability.
• Implement robust agent tool ecosystems: function-calling schemas, MCP (Model Context Protocol) server integrations, structured output validators, and fallback/retry logic — ensuring agents degrade gracefully when tools fail, APIs timeout, or LLM responses are malformed.
• Design and enforce agentic guardrails: input/output content filtering, prompt injection detection, loop-break conditions, human-in-the-loop approval gates, and cost circuit-breakers — balancing autonomous capability with the safety, auditability, and control requirements of enterprise deployments.

Qualifications

Required

• 8–10 years of hands-on AI / ML engineering experience with a demonstrable track record of delivering models and AI systems into production — not research or prototype environments.
• 2–3 years direct experience designing and shipping Agentic AI systems: multi-agent orchestration (LangGraph, AutoGen, CrewAI, or equivalent), tool-use / function-calling, agent memory architectures, and guardrail implementation at enterprise scale.
• Strong LLM engineering depth: RAG pipeline construction (chunking, embedding, hybrid search, re-ranking), prompt engineering at scale, fine-tuning with PEFT / LoRA, and multi-provider LLM integration (OpenAI, Anthropic Claude, Azure OpenAI, open-source Llama / Mistral).
• Production ML delivery: end-to-end ownership of model training, evaluation, deployment, and monitoring using MLflow, SageMaker, Azure ML, or Kubeflow; experience with data drift detection and automated retraining pipelines.
• Python 3.x mastery across the AI stack: PyTorch or TensorFlow, Hugging Face Transformers, LangChain / LlamaIndex, FastAPI for serving; comfort with async programming, streaming responses, and token-efficient prompt construction.
• Cloud AI platform experience on at least one of AWS (SageMaker, Bedrock), Azure (Azure OpenAI, Azure ML), or GCP (Vertex AI) — including containerised deployment on Kubernetes and infrastructure-as-code practices.
• AI safety and evaluation mindset: implemented content guardrails, output validation, evaluation frameworks (RAGAS, DeepEval, custom), and observability tooling (LangSmith, Arize, or equivalent) in production systems.

Preferred

• Experience with MCP (Model Context Protocol) server development for standardised agent tool integration.
• Knowledge of agentic memory systems: episodic memory with vector stores (Pinecone, Weaviate, pgvector), procedural memory for skill libraries, and semantic memory for knowledge graph integration.
• Exposure to AI governance frameworks: model cards, responsible AI checklists, bias audits, and compliance considerations for regulated industry deployments (HIPAA, GDPR, FDA AI/ML guidance).
• Familiarity with emerging multi-modal models (GPT-4o Vision, Gemini 1.5 Pro) and their application to document understanding, image-based agent perception, or speech-to-action pipelines.
• Research background or applied research publication in NLP, reinforcement learning, or agent systems.