Post Graduate Certificate in Agentic Systems & Production AI

Module 2: DevOps & Cloud Foundations for AI

• Devops Philosophy

• CI/CD Pipeline

• Git Workflows

• Linux CLI Essentials

• Cloud Service Models (AWS/Azure/GCP)

• Cloud Security Basics

• Cost Awareness

• Integrating ML Lifecycle Into CI/CD

Skills Acquired: DevOps philosophy | CI/CD pipelines | Git workflows | Linux CLI | AWS, Azure & GCP service models | Cloud security & cost basics

Learning Outcome: Understand CI/CD, Git workflows, Linux CLI and cloud service models, and how the ML lifecycle fits into them.

Module 3: Containerization & Orchestration For AI

• Docker Architecture

• Dockerfiles

• Container Networking

• Kubernetes Core Concepts (Pods, Services, Deployments)

• Autoscaling (HPA)

• GPU Scheduling Fundamentals For Inference Workloads

• Scaling ML Services

Skills Acquired: Docker & Dockerfiles | Container networking | Kubernetes Pods, Services, Deployments | HPA autoscaling | GPU scheduling for inference

Learning Outcome: Containerize an ML service and deploy it to Kubernetes with autoscaling and GPU-aware scheduling.

Module 4: Infrastructure as Code & Environment Automation

• IaC Principles

• Terraform Fundamentals

• Environment Reproducibility

• Secrets Management

• Infrastructure Versioning

• Provisioning ML-Ready Environments

• Advanced Terraform Modules & Workspaces

• Hybrid IaC + Kubernetes Integration

• Policy-as-Code & Compliance Automation

• Disaster Recovery & Scaling Automation

Skills Acquired: IaC principles | Terraform modules & workspaces | Secrets management | IaC + Kubernetes | Policy-as-code | DR & scaling automation

Learning Outcome: Provision reproducible ML environments with Terraform, applying secrets management, policy-as-code and recovery planning.

Module 5: ML Lifecycle & Pipeline Automation

• ML System Architecture

• Data Ingestion & Validation

• Feature Engineering Pipelines

• Feature Store Integration

• DAG Design

• Pipeline Orchestration Patterns

• Automation Checkpoints Before Deployment

• Reliable ML Pipeline Deployment

Skills Acquired: ML system architecture | Data ingestion & validation | Feature pipelines & feature store | DAG design | Orchestration patterns

Learning Outcome: Design an ML pipeline covering ingestion, validation, feature engineering and orchestration with pre-deployment checkpoints.

Module 6: Experiment Tracking & Reproducibility

• Experiment Tracking Principles

• Hyperparameter Logging

• Artifact Tracking

• Data & Model Versioning

• Reproducibility Workflows

• Governance Checkpoints In Experimentation

Skills Acquired: Experiment tracking | Hyperparameter & artifact logging | Data & model versioning | Reproducibility workflows | Governance checkpoints

Learning Outcome: Track experiments, artifacts and data/model versions so past results can be reproduced and reviewed.

Module 7: Model Packaging & Deployment

• Model Serialization (Pickle, ONNX)

• REST API Design For ML

• Batch Vs Real-Time Inference

• API Performance Testing

• Inference Latency Considerations

• Deployment Patterns (Containerized Serving), Scalability with Tools

Skills Acquired: Model serialization (Pickle, ONNX) | REST API design for ML | Batch vs real-time inference | API performance testing | Containerized serving

Learning Outcome: Package a model and expose it through a tested inference API, choosing between batch and real-time serving.

Module 8: CI/CD for ML Systems

• Continuous Training Pipelines

• Automated Validation Gates

• Model Approval Workflows

• Trigger-Based Deployment

• Rollback Strategies

• Integration With Orchestration Pipelines

Skills Acquired: Continuous training | Automated validation gates | Model approval workflows | Trigger-based deployment | Rollback strategies

Learning Outcome: Set up continuous training with automated validation gates, approval workflows and rollback paths.

Module 9: Monitoring, Drift & Governance

• Model Performance Monitoring

• Data Drift & Concept Drift Detection

• A/B Testing Strategies

• Retraining Triggers

• Governance Dashboards

• Interpreting Monitoring Metrics For Business Decisions

Skills Acquired: Performance monitoring | Data & concept drift detection | A/B testing | Retraining triggers | Governance dashboards

Learning Outcome: Monitor model performance, detect data and concept drift, and define retraining triggers and governance reporting.

Module 10: Model Registry & Enterprise Lifecycle

• Model Registry Design

• Approval Workflows

• Artifact Lineage

• Audit Trails

• Compliance Tracking

• Enterprise AI Lifecycle Management

Skills Acquired: Model registry design | Approval workflows | Artifact lineage | Audit trails | Compliance tracking | Enterprise AI lifecycle

Learning Outcome: Use a model registry to manage approvals, artifact lineage and audit trails across the model lifecycle.

Module 11: Enterprise LLM Hosting & Optimization

• LLM Serving Architectures (Managed APIs Vs Self-Hosted)

• Inference Optimization Basics

• VLLM Fundamentals

• Quantization Concepts (INT8/4-Bit Overview)

• GPU Utilization

• Throughput Benchmarking

• Latency Vs Cost Trade-Offs

• Production LLM Deployment Patterns

Skills Acquired: Managed vs self-hosted LLM serving | vLLM | Quantization (INT8/4-bit) | GPU utilization | Throughput benchmarking | Latency vs cost

Learning Outcome: Compare managed and self-hosted LLM serving, and benchmark throughput, latency and cost trade-offs.

Module 12: AI Agent Architecture & Foundations

• Fundamentals ofAI Agents and Architectures

• Agent Runtime Hosting, Service Isolation, Concurrency

• Docker/Kubernetes-Based Deployment, Observability, Audit Logging

• Operational Safety Controls

• Security and Compliance

• Resource Allocation and Infrastructure-Level Scaling of Enterprise Agent-Based Systems.

Skills Acquired: Agent architectures | Agent runtime hosting | Service isolation & concurrency | Docker/Kubernetes agent deployment | Observability & audit logging | Safety controls

Learning Outcome: Deploy agent runtimes on Docker and Kubernetes with service isolation, observability, audit logging and safety controls.

Module 13: Enterprise AI Hosting Strategies

• Advanced Inference Optimization

• Distributed LLM Serving

• Prompt Caching Strategies

• Enterprise-Grade APIs For Scalable

• High-Performance and Cost-Efficient Production Deployments.

Skills Acquired: Advanced inference optimization | Distributed LLM serving | Prompt caching | Enterprise-grade APIs | Cost-efficient deployment

Learning Outcome: Apply inference optimization, distributed serving and prompt caching to improve LLM performance and cost efficiency.

Module 14: LLM Cost Engineering & Secure Deployment

• Token Cost Modeling

• Caching Strategies

• Rate Limiting

• API Gateways

• Access Control

• Secure Model Endpoints

• Cost-Performance Dashboards

• Enterprise Proxy Patterns For LLM Usage, SLA/SLO Modeling

• Cost-Performance Trade-Offs

• High Availability Architectures

• Multi-Region Deployment Concepts

• Disaster Recovery Strategies

• Architecture Case Studies For ML & LLM Systems

Skills Acquired: Token cost modeling | Caching & rate limiting | API gateways & access control | Secure endpoints | SLA/SLO modeling | Multi-region HA & DR

Learning Outcome: Model token costs and secure LLM endpoints, and plan for SLAs/SLOs, high availability and disaster recovery.

Module 15: Vector Infrastructure & Retrieval Infrastructure

• Vector Database Architecture

• Indexing Strategies

• Sharding & Scaling

• Embedding Lifecycle Management

• Embedding Drift Detection

• Storage Tiering (Hot Vs Cold)

• Operational Considerations For Large-Scale Vector Systems

Skills Acquired: Vector DB architecture | Indexing strategies | Sharding & scaling | Embedding lifecycle | Embedding drift detection | Hot vs cold storage tiering

Learning Outcome: Select vector indexing, sharding and storage tiering approaches, and manage the embedding lifecycle and drift.

Module 16: RAG Ops

• Fundamentals of RAG

• Vector Databases and Indexing

• Secure Retrieval Workflows

• Monitoring and Observability

• RAG Governance

• RAGOps Integration

• Vector Database Governance

• Production-Ready RAG Case Studies

Skills Acquired: RAG fundamentals | Vector databases & indexing | Secure retrieval workflows | Monitoring & observability | RAG & vector DB governance | RAGOps integration

Learning Outcome: Build and operate a RAG pipeline with secure retrieval, monitoring and governance controls.

Module 17: AI Security & Compliance

• AI System Security and Robustness

• Compliance Frameworks

• Audit Logging

• Data Privacy for Secure, Compliant, and Resilient Enterprise AI Systems.

• Enterprise AI Systems Against Adversarial Threats While Ensuring Strict Alignment with Global Compliance Frameworks.

• Secure Deployment

• Audit Governance

• Privacy‑Preserving Practices

Skills Acquired: AI system security & robustness | Adversarial threat defense | Compliance frameworks | Audit logging | Data privacy & privacy-preserving practices

Learning Outcome: Apply AI security, audit logging and data privacy practices aligned to compliance frameworks.

Capstone: End-To-End Enterprise ModelOps and LLM Infrastructure Capstone

• Integrating CI/CD Automation

• Model Registry

• Monitoring

• Drift Detection

• Serving Optimization

• Cost Governance

• Architecture Documentation

• Enterprise Deployment Review

Skills Acquired: CI/CD automation | Model registry | Monitoring & drift detection | Serving optimization | Cost governance | Architecture documentation | Deployment review

Learning Outcome: Deliver an integrated ModelOps and LLM infrastructure project with architecture documentation and a deployment review.

This Production AI course is designed for software and backend engineers, DevOps and cloud professionals, platform engineers, Site Reliability Engineers, AI and ML professionals, data professionals, solution architects, tech leads and engineering managers who want to develop capabilities for taking AI systems into enterprise production.

Applicants must hold a bachelor’s degree or diploma and have at least three years of work experience. Participants should also have:

• Intermediate proficiency in Python

• Basic knowledge of cloud infrastructure

• Basic knowledge of DevOps practices

The Production AI course curriculum covers the systems and practices required to manage models, LLMs, RAG systems and AI agents in production. The 24-week curriculum spans AI and ML foundations, DevOps and cloud foundations, Docker, Kubernetes, infrastructure as code, production ML pipelines, MLOps, experiment tracking, model registries, model deployment, monitoring and drift detection. It also covers enterprise LLM hosting and optimisation, agentic systems, agent runtime deployment, vector infrastructure, RAGOps, AI security, governance and compliance, culminating in an Enterprise ModelOps and LLM Infrastructure Capstone.

This Production AI programme extends beyond traditional MLOps by covering the production requirements of ML models, LLMs, RAG systems and AI agents.In addition to ML pipelines, deployment and model monitoring, participants explore LLM hosting and cost engineering, vector infrastructure, RAGOps, agent runtime deployment, observability, security and governance.

Yes. This programme includes dedicated coverage of LLMOps, AgentOps and RAGOps, alongside MLOps and cloud-native AI infrastructure.Participants explore:

• LLMOps: LLM hosting, vLLM, quantisation, benchmarking, prompt caching and cost management

• AgentOps: Agent runtime hosting, service isolation, concurrency, observability and audit logging

• RAGOps: Vector infrastructure, indexing, retrieval workflows, monitoring and governance

Yes. The programme covers enterprise AI infrastructure engineering across containers, orchestration, infrastructure automation, scalable serving and reliability.Participants learn about Docker, Kubernetes, Terraform, GPU-aware scheduling, autoscaling, high availability and disaster recovery concepts for Production AI environments.

Participants gain practical exposure to 25+ tools and platforms used across AI infrastructure, MLOps, LLMOps, RAGOps, monitoring and agentic systems. The indicative toolset includes AWS, Docker, Kubernetes, Terraform, Apache Airflow, MLflow, FastAPI, ONNX Runtime, KServe, vLLM, Hugging Face, LangGraph, Prometheus, Grafana and Evidently AI.

The programme runs for six months, is delivered live online, and requires approximately 8–10 hours of weekly effort. The programme fee is INR 1,50,000 plus applicable GST.

Participants who successfully meet the completion requirements will receive an e-Certificate from iHUB DivyaSampark, IIT Roorkee. Participants must maintain at least 50% attendance in live sessions and achieve a minimum overall score of 50% across the required assessments.

An optional two-day campus immersion at IIT Roorkee’s Noida campus is available upon programme completion.