Advanced Certificate Programme in Agentic AI and RAG Engineering

• Python—APIs, concurrency, and Pydantic

• File handling, logging, config, secrets

• LLM-assisted coding workflows

• FastAPI—streaming, background tasks, and validation

• Logging and tracing hooks

• UI integration

• Testing Vs Evaluation distinction

• LLM APIs - Tokens, cost, and latency

• Structured outputs and tool-calling

• Streaming, retries, and timeouts

• Model selection - closed Vs open-weights

• Local model integration

• Structured, tool-aware, and few-shot prompting

• Evaluating non-deterministic outputs

• LLMs-as-judge

• Pairwise comparison

• Rubric-based evaluations

• Golden dataset construction

• Critic-Creator prompt loop

• Use-case discovery exercise

• Why RAG - and when not to use it

• Build a RAG pipeline from first principles

• Embeddings and similarity by hand

• KPI tracking begins

• Embedding model selection

• Vector database trade-offs

• Similarity metrics

• Index structures

• Multi-format ingestion—PDF, HTML, Docx

• Chunking strategies

• Metadata enrichment

• PII detection and compliance

• Hybrid search

• Reranking

• Query rewriting and expansion

• Multi-query retrieval

• Retrieval framework integration

• Prompt and context compression

• Multi-layer caching strategies

• Knowledge base lifecycle management

• Versioning and drift handling

• Retrieval and generation quality metrics

• Evaluation frameworks

• Evaluation as a code

• Failure taxonomy for RAG systems

• Systematic debugging workflow

• Trace reading

• Cost analysis

• Function calling — design and implementation

• Tool reliability principles

• API-calling agents

• Text-to-SQL

• Web search integration

• Data operations and audit logging

• Build a working agent from scratch

• Agent control loops and safety guards

• Failure-Oriented Design — failure types and defenses

• Testing agents — unit, integration, mocking

• Memory types and selection

• When not to add memory

• Retrieval and forgetting strategies

• State machine design for agents

• Deterministic Vs agentic routing

• Guardrail node patterns

• Parallel workflow execution

• Planner-executor patterns

• Reflection and self-correction

• Streaming agent UX

• Human oversight — approval gates, escalation, audit trails

• Retrieval-as-a-tool; Grounded agents with citation

• Prompt injection — types and defenses

• Data exfiltration prevention

• Multi-source retrieval orchestration

• When NOT to use multi-agent

• Cost and complexity trade-offs

• Single-agent alternatives

• Decision framework

• Supervisor-worker, planner-executor, debate patterns

• Role and tool partitioning

• Orchestration frameworks

• Shared state and message passing

• Delegation and aggregation

• Deadlocks, cascade failures, circular delegation

• Coordination primitives

• Cross-agent tracing

• Non-deterministic bug reproduction

• Cost explosion diagnosis

• Load testing

• MCP as an emerging integration standard

• Client/server architecture

• Transport and security model

• When to use MCP vs direct tool-calling

• Build and expose a custom MCP server

• Agent-to-MCP integration

• Trade-off analysis

• Logging, tracing, and metrics for AI

• Hands-on instrumentation

• Cross-system tracing

• Observability as a failure-detection surface

• Online vs offline evaluation

• Prompt and model versioning

• Regression testing

• Shadow deployments and A/B testing

• Cost breakdown by component

• Model routing and cascading

• Budget controls

• Queue systems

• Local model deployment

• SLA/SLO definition

• Hallucination mitigation

• AI guardrails

• Security consolidation

• DPDP Act, GDPR

• Audit trails and data residency

• Containerisation for AI

• Prompt, model, and dataset versioning

• CI/CD pipelines with eval gates

• Deployment strategies

• Production operations

• Production system presentations

• Reading AI research critically

• Agent benchmark literacy

• Career next steps

• Introduction to Vector Databases and Chroma DB

• Vector Databases for Recommendation Systems and RAG

• RAG Framework

• Prompt Engineering and LangChain

This programme is designed for technology professionals who want to deepen their expertise in modern AI system design and deployment. It is particularly relevant for software engineers, AI/ML engineers, data engineers, backend developers, solution architects, technical product managers, platform leads, and innovation professionals.

The curriculum is best suited for individuals who already have some exposure to programming and want to move beyond experimentation with AI tools to building production-ready systems. Professionals interested in multi-agent systems, enterprise AI applications, retrieval systems, and AI engineering workflows will find the programme especially valuable.

Whether your goal is to transition into AI systems engineering or strengthen your ability to design and deploy AI solutions within your organisation, the programme provides a structured pathway to build these capabilities.

Organisations are rapidly moving beyond standalone AI tools and chatbots towards intelligent systems that can reason, retrieve information, use external tools, and execute tasks autonomously. These systems increasingly combine Agentic AI and RAG architectures to improve reliability, accuracy, and automation.

As AI adoption accelerates across industries, professionals who understand how to build and deploy these systems are becoming increasingly valuable. Businesses are looking for individuals who can bridge the gap between AI models and real-world applications by creating scalable, production-ready solutions.

Learning Agentic AI and RAG Engineering helps professionals develop practical skills that are directly applicable to the next generation of enterprise AI systems and positions them to contribute to high-impact AI initiatives.

Many Generative AI programmes focus primarily on prompt engineering, model usage, or AI-powered productivity tools. While these skills are important, they represent only one part of the AI ecosystem.

The Advanced Certificate Programme in Agentic AI & RAG Engineering goes significantly deeper by focusing on how AI systems are designed, orchestrated, deployed, and governed. Learners explore topics such as multi-agent architectures, retrieval systems, memory design, orchestration frameworks, evaluation, observability, deployment, and AI governance.

A key differentiator is the programme's combined focus on both Agentic AI and RAG Engineering. This reflects how modern enterprise AI applications are increasingly built and gives learners a more comprehensive understanding of AI system design than programmes focused on a single area.

The Agentic AI and RAG certification programme covers the complete lifecycle of building modern AI systems. Participants begin by strengthening their understanding of AI system design fundamentals before progressing to advanced topics in Agentic AI, RAG Engineering, multi-agent orchestration, deployment, and governance.

Learners will explore how agents reason and make decisions, how retrieval systems improve AI performance, how memory can be incorporated into AI workflows, and how AI services can be deployed and monitored in production environments.

The curriculum also includes practical exposure to evaluation frameworks, security considerations, observability tools, and AI governance concepts. By the end of the programme, learners will have the knowledge and experience required to design and deploy production-ready AI applications.

The programme provides hands-on exposure to more than 15 industry-relevant tools and frameworks used in modern AI engineering workflows.

Participants will work with technologies such as LangGraph, Tavily, FastAPI, Docker, Qdrant, vector databases, and evaluation frameworks commonly used to build Agentic AI and RAG systems. These tools help learners understand how different components of an AI system work together—from orchestration and retrieval to deployment and monitoring.

Rather than focusing solely on theory, the programme emphasises practical implementation, enabling learners to apply concepts through assignments, projects, and the capstone project using tools that are increasingly being adopted across industry.

Yes. Practical application is a central component of the AI and RAG programme by IITM Pravartak. Learners complete more than 20 assignments and projects that are designed to simulate real-world AI engineering challenges.

These projects cover a wide range of topics, including multi-agent workflows, retrieval systems, memory architectures, orchestration frameworks, tool integration, and deployment. Participants are encouraged to apply concepts throughout the programme rather than waiting until the end to build solutions.

The programme culminates in a capstone project that allows learners to bring together concepts from across the curriculum and demonstrate their ability to design and deploy a complete AI system.

Yes. The AI and RAG certificate course by IITM Pravartak is an engineering-focused programme and prior programming experience is recommended.

Learners should be comfortable with basic programming concepts and ideally have some familiarity with Python. Experience working with APIs, backend systems, Git/GitHub, VS Code, or Linux environments can also be beneficial.

While the programme reinforces foundational concepts where necessary, it is designed for professionals who want to build and deploy AI systems rather than individuals seeking a purely introductory or no-code AI programme.

Participants who successfully complete the programme will receive a certificate from IITM Pravartak, recognising their achievement and commitment to building expertise in Agentic AI and RAG Engineering.

In addition to the IITM Pravartak certificate, the programme also includes IBM credentials that further strengthen learners' professional profiles. These credentials can help demonstrate proficiency in relevant AI concepts and technologies and serve as valuable additions to professional portfolios.

The combination of practical skills, project work, and recognised credentials helps learners showcase their capabilities in a rapidly evolving AI landscape.

Yes. The programme includes an optional campus immersion experience at IIT Madras Research Park.

This immersion provides participants with an opportunity to engage with the broader innovation ecosystem, interact with fellow professionals, and gain exposure to an environment that supports research, entrepreneurship, and technology development.

For many learners, the immersion serves as a valuable networking opportunity and offers additional perspective on how advanced technologies are being explored and applied in real-world settings.

Yes. The programme covers deployment stacks such as FastAPI and Docker, along with observability, cost control, evaluation frameworks, CI/CD practices, and secure AI system design to ensure scalability and reliability.

Yes. You will learn multi-agent architectures including role separation, planner–executor patterns, shared memory coordination, tool orchestration, and performance validation across concurrent agents.

Yes, there are certain system requirements to ensure a seamless programme experience:

• Laptop/desktop with Windows 10/11 (64-bit) or Ubuntu 20.04+ or macOS 12 or above, minimum 8 GB RAM (16 GB recommended)

• 4-core Intel i5/i7 or AMD equivalent or Apple Silicon (M1/M2/M3 or above),

• 120 GB free SSD storage for windows/linux or 30GB free disk space for MacOS

• Ability to install Python, Docker, and VM

• Stable internet connection

Note:

• Learners must have a personal system meeting minimum hardware and virtualisation requirements to run hands-on labs. Company-issued, restricted, or low-spec devices (e.g., 8 GB RAM) may not support all lab activities.