🧩 AI, RAG & Agentic Systems

25 Days × 2 Hrs (50 Hrs) | 6 Modules | Free-tier-first stack | Three daily batches on one codebase

From tokens to a deployed, evaluated, guard-railed multi-agent system — over the learner's own corpus, chosen on Day 1 and carried to Day 25.

🔍 Retrieval That Works

Chunking, vector stores, hybrid search, reranking

BM25 + DenseRRFCross-Encoder

📊 Evidence, Not Claims

Golden sets, held-out splits, build-failing thresholds

RAGASLLM-as-Judgepytest Gates

🕸️ Graph & Agentic Retrieval

GraphRAG, KAG, retrieval that corrects itself

NetworkXNeo4jSelf-Correction

🛡️ Bounded Agents

Allowlists, step caps, spend caps, injection defence

LangGraphTracesRed-Team

🪜 The Build Ladder — one corpus, twenty-five days

Each module's output is the next module's input. There is no standalone week.

rag_v1
Day 8

➔

Failure Audit
Day 10

➔

rag_v3
Day 12

➔

Eval Suite
Day 14

➔

Deployed API
Day 15

➔

Guarded Agent
Day 21

➔

Capstone
Day 25

📚 Module Map

Module	Title	Days	Hrs	Hands it forward
Module 0	Orientation	1	2	A working environment and a chosen corpus
Module 1	Foundations — how models actually work	2–4	6	Tokens, embeddings, structured output
Module 2	The RAG pipeline, end to end	5–9	10	A tagged rag_v1 baseline
Module 3	Diagnose, improve, prove	10–14	10	An eval suite with thresholds
Module 4	Production, graphs and self-correction	15–18	8	A deployed service and self-judging retrieval
Module 5	Agents	19–23	10	A traced, guarded multi-agent pipeline
Module 6	Context, evaluation and capstone	24–25	4	An honestly measured system, defended

🚦 Four Hard Gates

Each is a test in the lab repository, so gate status is a fact rather than an opinion.

Day	Gate	Test	What it proves
1	Environment	`test_day00_setup.py`	Runs from a fresh clone on someone else's machine
8	rag_v1 baseline	`test_day08_rag_v1.py`	A reproducible baseline to measure everything against
14	Eval threshold	`test_day14_eval_gate.py`	Quality regressions fail the build
21	Agent isolation	`test_day21_isolation.py`	The agent reaches only what it was given

🗂️ Track Assets

Course Structure

13-sheet workbook — syllabus, labs, rubric, risks, production tracker

DRISYON_AI_RAG_Agentic_50H_Course_Structure.xlsx

Facilitator Decks

25 decks, 300 slides, plus 25 printable run-sheets

decks/decks/facilitator_notes/

Lab Repository

25 day folders, shared modules, 4 pytest gates, sample corpus

labs/ai-rag-agents-labs.zip

NotebookLM Kits

Student study guide, trainer playbook, director strategy & rubric

notebooklm_kits/

Three batches daily (IST): Corporates 07:00–09:00 · Students 16:30–18:30 · Employees 19:30–21:30. One deck, one lab repository, one marking scale — the corpus and the capstone framing are what differ.

Distinct from the Agentic AI track, which is a separate, shorter programme for a different audience.