Retrieval-Augmented Generation · 13 weeks

Build AI applications that know how to use your data.

Go beyond prompting. Learn to build Retrieval-Augmented Generation (RAG) applications that retrieve relevant information, answer with evidence and can be tested, improved and deployed.

● ● ●　 FDE / KNOWLEDGE COPILOT ● Connected

COMPANY KNOWLEDGE COPILOT

How does our travel policy handle international reimbursements?

✓ Answer grounded in sourcesThe policy specifies the eligible expense categories and approval process. The exact limits should be checked against the applicable policy section.

▤ Travel Policy · §4.2▤ Handbook · p.18↗ View evidence

Duration13 weeks

StartsNovember 2026

Commitment8–10 hrs a week

FormatLive, online

Enrol now Read the week-by-week curriculum See all 8 projects ↓Applications close 30 October 2026

LEARN FROM PRIMARY SOURCESOriginal RAG researchMicrosoft ArchitectureAnthropic EngineeringGoogle CloudRagas Evaluation

Start with the idea

What is RAG, and why does it change what you can build?

Retrieval-Augmented Generation combines a retrieval system with a language model. When a user asks a question, the application searches a knowledge source, selects relevant passages and gives them to the model as context for its response.

That lets an application work with selected documents, internal knowledge and changing information instead of relying only on what the model learned during training.

RAG does not guarantee truth. Source quality, retrieval relevance, permissions and evaluation all matter.

Read the original research ↗

Inside a RAG application

A simple view of the query-time workflow

▤

1. Your knowledgePDFs, guides, web pages and structured data

⌕

2. RetrieveSearch for relevant chunks and evidence

✳

3. AugmentProvide question and context to the LLM

✓

4. GenerateAnswer with sources, or abstain if evidence is weak

Prepare knowledge: parse → chunk → embed → indexThen retrieve at question time ↗

Why this architecture

A convincing answer is not enough. It needs evidence.

Imagine asking about a company's travel policy. A generic model may produce a plausible response, but without the actual policy it cannot verify the organisation's rule.

✕ General-purpose response

No connected company knowledge

USER ASKS“What is the international travel reimbursement limit?”

POTENTIAL FAILURE“Most companies reimburse reasonable travel expenses…”
Plausible, but not evidence of this company's actual policy.

✕ May answer generically or infer details
✕ May lack current internal information
✕ No verifiable source unless one is supplied

✓ RAG-powered response

Connected to approved knowledge

RETRIEVED EVIDENCETravel Policy · Section 4.2
“International expenses must follow the limits and approval process defined in this section.”

GROUNDED RESPONSE“Here is the applicable policy section. I can summarise its stated limit and approval steps.”
The application can cite the actual source passage.

✓ Uses external knowledge at answer time
✓ Can show evidence and source references
✓ Can be tested against known questions and documents

Illustrative scenario, not a real company's policy. RAG can still fail when retrieval misses evidence or the model misuses it.

Use cases

What can you actually build with RAG?

RAG is useful when an application needs to answer questions over a selected body of knowledge. Here are concrete starting points.

▤

Enterprise knowledge assistant

Search internal handbooks, process notes, product policies and operating procedures.

Example: “Which onboarding steps must be completed before account activation?”

☏

Customer support copilot

Retrieve answers from approved product documentation and troubleshooting guides.

Example: “What are the verified steps to resolve this error code?”

⌕

Research & document analysis

Search long reports, compare passages and produce evidence-linked summaries.

Example: “Compare assumptions stated in these three reports.”

⬡

Compliance & policy search

Locate relevant policy passages for review by qualified people.

Example: “Which section describes the approval requirements?”

⌘

Developer documentation copilot

Navigate API references, setup instructions and technical guides.

Example: “Which parameters are required for this endpoint?”

▦

Personal knowledge assistant

Ask questions across a curated library of notes, PDFs and meeting summaries.

Example: “What decisions did we record about this project?”

The engineering opportunity

Move from using AI tools to engineering AI applications.

Knowing how to prompt a model is a starting point. Building a useful AI system requires data preparation, retrieval, evaluation, security, deployment and operational trade-offs.

Connect AI to useful knowledge

Understand ingestion, chunking, embeddings, indexing and retrieval.

Debug instead of guessing

Inspect retrieved passages and trace why an answer is weak.

Measure quality systematically

Build test sets and compare retrieval and response quality.

Ship a demonstrable application

Connect the interface and backend, deploy a beta and explain the architecture.

AI skills are appearing in job postings

US AI job postings citing generative-AI skills, 2023 vs 2024

Source: Stanford HAI AI Index 2025, citing Lightcast. US AI postings and generative-AI skills overall, not RAG-specific roles or Indian hiring data. Context, not a job guarantee.

Read the AI Index report ↗

Proof over promises

Build a portfolio that shows how you think, not just what you watched.

A credible AI project lets someone inspect the code, reproduce the result, understand the design and see how you tested the application.

Working application

A runnable project with a clear user problem, setup instructions and usable interface.

Readable code

A repository with understandable modules, configuration and documented assumptions.

Evaluation evidence

A repeatable question set, expected evidence, test results and regression checks.

Architecture explanation

A diagram showing ingestion, retrieval, generation, data sources and boundaries.

Deployment and demo

A demonstrable application, known limitations and reproducible run instructions.

Trade-off thinking

An explanation of quality, latency, cost, access control and design decisions.

These are portfolio deliverables the program is designed around. Career outcomes depend on individual experience, effort, project quality and employer requirements.

The learning path

13 weeks. 26 classes. One coherent engineering path.

Each 150-minute class combines concept instruction, an instructor demonstration, guided lab work, independent implementation and a submission or assessment.

01WEEKS 1–2 · CLASSES 1–4

AI foundations

LLMs, APIs, structured output, Python and document processing.

02WEEKS 3–4 · CLASSES 5–8

Core RAG

Embeddings, vector search, retrieval pipelines and citations.

03WEEKS 5–7 · CLASSES 9–14

Advanced RAG

Chunking, multi-source retrieval, optimisation and evaluation.

04WEEKS 8–10 · CLASSES 15–20

Real-world systems

Hybrid search, tools, security, reliability and cost.

05WEEKS 11–13 · CLASSES 21–26

Deploy & defend

Backend, frontend, testing, release and capstone defence.

01AI application foundationsWeek 1 · From model concepts to a structured assistant

Saturday · Class 1

LLMs and AI application architecture

Understand tokens, context windows, model inputs and outputs, API applications and failure modes.

Sunday · Class 2

API calls, prompting and structured output

Call a model, design instructions, validate structured responses and handle malformed outputs.

Weekly outcome: Mini-project 1 — Structured AI Assistant.

02Python and document intelligenceWeek 2 · Turn raw files into usable information

Saturday · Class 3

Python foundations and document processing

Work with files, text extraction, data structures and reusable processing functions.

Sunday · Class 4

Document intelligence and extraction

Extract selected fields from sample documents and validate structured data.

Weekly outcome: Mini-project 2 — Document Intelligence Extractor.

03Embeddings and semantic searchWeek 3 · Retrieve by meaning, not only exact words

Saturday · Class 5

Embeddings and semantic similarity

Understand vector representations, similarity measures and practical limitations.

Sunday · Class 6

Vector search and vector databases

Create an index, retrieve ranked results and inspect matching passages.

Weekly outcome: Mini-project 3 — Semantic Search Engine.

04Build your first RAG applicationWeek 4 · From retrieval to source-grounded answers

Saturday · Class 7

RAG architecture and retrieval pipeline

Connect ingestion, retrieval, prompt construction and generation.

Sunday · Class 8

Citation-based question answering

Return answers with source references and test missing-evidence behaviour.

Weekly outcome: Mini-project 4 — Citation-Based RAG Assistant.

05Chunking and retrieval optimisationWeek 5 · Understand why information gets missed

Saturday · Class 9

Chunking strategies and metadata

Compare fixed-size and structure-aware chunking, overlap and metadata.

Sunday · Class 10

Retrieval debugging and optimisation

Inspect failed queries, adjust retrieval settings and compare results.

Weekly outcome: Mini-project 5 — RAG Quality Optimiser.

06Frameworks and multi-source knowledgeWeek 6 · Build a copilot across data types

Saturday · Class 11

RAG frameworks and orchestration

Understand pipeline abstractions and when explicit code is clearer.

Sunday · Class 12

Multi-source and structured-data RAG

Combine documents with a structured data source and retain attribution.

Weekly outcome: Mini-project 6 — Multi-Source Knowledge Copilot.

07RAG evaluation and benchmarksWeek 7 · Measure quality beyond a good demo

Saturday · Class 13

Evaluation design and benchmark datasets

Write representative questions, expected answers and evidence references.

Sunday · Class 14

Automated evaluation and regression tests

Score runs, compare changes and prevent known failures from returning.

Weekly outcome: Mini-project 7 — RAG Evaluation Lab.

08Hybrid search and complex queriesWeek 8 · Improve retrieval for real-world questions

Saturday · Class 15

Hybrid search and reranking

Combine keyword and semantic retrieval, rank fusion and reranking.

Sunday · Class 16

Query rewriting and complex documents

Improve ambiguous queries and reason about tables, long documents and multi-step information needs.

Weekly outcome: Capstone retrieval strategy documented and benchmarked.

09Function calling and tool-enabled workflowsWeek 9 · Move from answering to controlled actions

Saturday · Class 17

Function calling and tool interfaces

Define tools, schemas, validation, execution boundaries and failure messages.

Sunday · Class 18

Tool-enabled retrieval workflows

Combine retrieval with a controlled lookup or action and log execution.

Weekly outcome: Mini-project 8 — Tool-Enabled AI Copilot.

10Security, reliability and costWeek 10 · Think beyond the happy path

Saturday · Class 19

Access control, prompt injection and security

Explore permission boundaries, untrusted content, data exposure and adversarial tests.

Sunday · Class 20

Observability, reliability and cost

Log traces, handle timeouts, estimate usage and identify bottlenecks.

Weekly outcome: Capstone security checklist, observability plan and cost assumptions.

11Backend, frontend and deployed betaWeek 11 · Turn components into a usable product

Saturday · Class 21

Backend API and application architecture

Expose retrieval through an API and organise application modules.

Sunday · Class 22

Frontend integration and deployed beta

Connect the interface, show answers and sources, and deploy a beta.

Weekly outcome: Capstone beta with end-to-end question answering.

12Testing, code review and release readinessWeek 12 · Make the application reproducible

Saturday · Class 23

End-to-end tests and regression checks

Test ingestion, retrieval, citations, failure paths, permissions and outputs.

Sunday · Class 24

Code review, release readiness and documentation

Review maintainability, configuration, setup steps, limitations and release checks.

Weekly outcome: Release candidate with tests, documentation and evaluation report.

13Capstone finalisation and architecture defenceWeek 13 · Present the product and explain trade-offs

Saturday · Class 25

Capstone finalisation and release review

Resolve release blockers, rerun acceptance tests and prepare the demonstration.

Sunday · Class 26

Final demonstration and architecture defence

Demonstrate the application, explain architecture choices and discuss evaluation and limitations.

Weekly outcome: Deployed Enterprise Knowledge Copilot, repository, documentation and technical defence.

Build your portfolio

Eight mini-projects. Eight distinct capabilities.

All eight projects are listed here. Expand each project to see skills and acceptance criteria.

PROJECT 01Week 1

Structured AI Assistant

Build an assistant that returns validated, predictable structured outputs instead of unconstrained prose.

Skills & acceptance criteria

Skills: API integration · Prompt design · Schema validation · Error handling

Acceptance: Implement the workflow, handle at least one failure case, document setup and demonstrate the result.

Portfolio deliverable
Runnable assistant, output schema, validation rules and tests.

PROJECT 02Week 2

Document Intelligence Extractor

Extract selected fields from sample PDFs and documents, then validate resulting records.

Skills & acceptance criteria

Skills: File handling · Text extraction · Structured data · Validation

Acceptance: Implement the workflow, handle at least one failure case, document setup and demonstrate the result.

Portfolio deliverable
Sample corpus, extraction pipeline, structured output and failure-case tests.

PROJECT 03Week 3

Semantic Search Engine

Search a knowledge collection by semantic similarity and inspect ranked passages.

Skills & acceptance criteria

Skills: Embeddings · Similarity · Vector indexing · Retrieval

Acceptance: Implement the workflow, handle at least one failure case, document setup and demonstrate the result.

Portfolio deliverable
Indexed corpus, search interface and retrieval experiment notes.

PROJECT 04Week 4

Citation-Based RAG Assistant

Connect retrieval to generation and return answers with references to supporting passages.

Skills & acceptance criteria

Skills: RAG pipeline · Context · Citations · Abstention

Acceptance: Implement the workflow, handle at least one failure case, document setup and demonstrate the result.

Portfolio deliverable
End-to-end Q&A app, source references and missing-evidence tests.

PROJECT 05Week 5

RAG Quality Optimiser

Run controlled experiments with chunking, top-k and retrieval settings.

Skills & acceptance criteria

Skills: Chunking · Debugging · Experiment design · Metrics

Acceptance: Implement the workflow, handle at least one failure case, document setup and demonstrate the result.

Portfolio deliverable
Evaluation questions, configuration comparisons and documented findings.

PROJECT 06Week 6

Multi-Source Knowledge Copilot

Combine document collections and a structured data source while preserving attribution.

Skills & acceptance criteria

Skills: Multi-source ingestion · Orchestration · Metadata · Source tracking

Acceptance: Implement the workflow, handle at least one failure case, document setup and demonstrate the result.

Portfolio deliverable
Integrated sources, source-aware responses and architecture notes.

PROJECT 07Week 7

RAG Evaluation Lab

Create a repeatable benchmark to evaluate retrieved context and generated answers.

Skills & acceptance criteria

Skills: Benchmark design · Faithfulness · Precision/recall · Regression

Acceptance: Implement the workflow, handle at least one failure case, document setup and demonstrate the result.

Portfolio deliverable
Test dataset, evaluation runner, score report and regression checks.

PROJECT 08Week 9

Tool-Enabled AI Copilot

Extend an assistant with a controlled tool workflow, validated inputs and observable execution.

Skills & acceptance criteria

Skills: Function calling · Tool schemas · Validation · Safety boundaries

Acceptance: Implement the workflow, handle at least one failure case, document setup and demonstrate the result.

Portfolio deliverable
Tool-enabled workflow, execution logs and failure-path tests.

Illustrative interface mockups. Repositories and live demos will be linked once verified.

The integrated capstone

Enterprise Knowledge Copilot

Build one coherent application from ingestion to deployment. Integrate the capabilities developed across the program into a system you can run, test, demonstrate and explain.

Document and data ingestionGrounded answers with citationsStructured data integrationControlled tool workflowEvaluation and regression testsSecurity and access checksDeployment and monitoring planArchitecture defence

See the weekly outcomes ↑

The RAG reading room

Learn from the people building the field.

A serious engineering program should point learners to original research, technical architecture guidance and evaluation frameworks, not ask them to rely on marketing claims alone.

Original research · 2020

Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

The foundational paper describing a retriever-generator combination for knowledge-intensive tasks.

Read the paper ↗

Anthropic Engineering · 2024

Introducing Contextual Retrieval

Explores how chunk context affects retrieval and methods to improve retrieval results.

Read the article ↗

Microsoft Architecture Center

Design and develop a RAG solution

Architecture guidance covering retrieval choices, evaluation and a structured RAG development approach.

Read the guide ↗

Google Cloud

Retrieval-Augmented Generation

A practical overview of connecting generative models with external knowledge sources.

Explore the overview ↗

IBM Think

What is retrieval-augmented generation?

A conceptual introduction to RAG, use cases and the relationship between retrieval and model knowledge.

Read the explainer ↗

Open-source evaluation

Ragas: Evaluation metrics

Explore metrics for context precision, context recall, faithfulness and response relevance.

Explore the metrics ↗

Stanford HAI · 2025

AI Index Report 2025

Evidence on AI research, adoption and labour-market indicators. Interpret each figure by geography and scope.

Explore the report ↗

Microsoft evaluation guide

RAG end-to-end evaluation

Why quality must be assessed across retrieval and generation with documented experiments and multiple metrics.

Read the guide ↗

Practice

Reproduce, compare and document

Use research as a starting point. Run small experiments and record what works for your data and requirements.

See where it fits ↑

FDE School is not affiliated with these publishers. Links are independent learning resources and may change over time.

Designed for different starting points

Who is this program for?

The program starts with foundations and progresses toward application engineering. Learners should be ready to practise between classes and work through coding exercises.

⌘

Beginners to RAG

Build an understanding of LLM applications and learn retrieval step by step.

▣

Working professionals

Add practical AI application skills to current technical or product experience.

〈/〉

Developers

Extend development skills into retrieval, evaluation and LLM-powered apps.

↗

Product builders

Turn a use case into a testable prototype with explainable architecture.

Fee and dates

What it costs and when it starts.

Fee₹44,999+ 18% GST

Total with GST₹53,099EMI on major credit cards and via NBFC partners

Seat booking₹5,000Registration fee that books your seat

StartsNovember 2026Applications close 30 October 2026

Enrol now

Before you enrol

Questions worth asking before you invest.

Clear answers help you decide whether the time commitment and learning outcomes fit your goals.

Do I need prior RAG experience?

No prior RAG experience is required. The curriculum starts with AI application foundations and progresses through retrieval, evaluation and deployment. Basic programming familiarity will help with coding labs.

What exactly will I build?

Eight mini-projects cover structured outputs, document extraction, semantic search, citation-based RAG, retrieval optimisation, multi-source retrieval, evaluation and tool-enabled workflows, followed by an integrated capstone.

What is the fee?

₹44,999 + 18% GST (₹53,099 total). A ₹5,000 registration fee books your seat. EMI is available on major credit cards and through NBFC partners.

Who should apply, and is the program selective?

Freshers and working professionals from technical or analytical backgrounds. A Bachelor’s (CS/IT/Engineering/Maths or related) is expected. Yes, it is selective: apply online, take a pre-screening test, receive an offer letter, then pay to begin.

Will I have a portfolio at the end?

The program is designed around project deliverables, code, tests, evaluation reports and a capstone. Portfolio quality depends on the work and documentation you complete.

How will my work be assessed?

Through mini-projects, capstone milestones, testing and evaluation, the final application and an individual technical defence.

How much time should I set aside?

There are 26 live classes of 150 minutes each, plus independent implementation, debugging and capstone work. Plan on 8–10 hours a week in total.

Does this guarantee a job or salary increase?

No. The program is designed to help you develop and demonstrate practical skills. Career outcomes depend on your background, effort, project quality and employer requirements.

Does the certificate guarantee industry recognition?

The certificate is issued by FDE School and documents completion under its criteria. It is not third-party accreditation.

Learn · Build · Evaluate · Deploy

Turn your curiosity about AI into something you can demonstrate.

Explore the full learning path, inspect every project and decide whether the RAG course fits your goals.

Program details reflect the current curriculum plan. Career outcomes depend on your background, effort and employer requirements.

Enrol now Review curriculum ↑Talk to admissions ↗