Practical Agentic AI: RAG, Planning & Vector Search

This course contains the use of artificial intelligence. Generative AI is moving from reactive “knowers” to proactive “doers” that perceive, plan, and act toward goals. This shift—Agentic AI—pairs LLM reasoning with tools, memory, and workflows so systems can execute multi-step tasks autonomously.

Enterprises now expect agents that ground answers with RAG, orchestrate APIs, and operate reliably with guardrails—raising new questions about autonomy, accountability, and oversight. What This Course CoversYou’ll learn an end-to-end Agentic AI stack: the Perceive→Reason→Act loop; Retrieval-Augmented Generation; planning & memory; the MCP (Model–Controller–Prompter) workflow; and framework choices (LangChain, LlamaIndex, CrewAI, AutoGen). We translate concepts into an applied build: a CLI “Personalized News Curator” that uses Tavily for live search, ChatGPT/Gemini for ranking & summaries, an in-memory/SQLite → ChromaDB store, topic-pillar weighting, semantic re-ranking, and explanation generation.

What You Will LearnDifferentiate LLMs vs. Agentic AI across autonomy, memory, and tool use. Apply the Perceive→Reason→Act loop to real tasks.

Implement MCP (Model–Controller–Prompter) orchestration for agents. Ground responses with RAG for factuality and reliability. Build a CLI agent that collects preferences and runs a continuous recommendation loop.

Integrate Tavily search + ChatGPT/Gemini for retrieval and ranking. Persist interaction history (SQLite) and migrate to ChromaDB embeddings. Engineer topic “pillars,” weighted selection, and semantic re-ranking.

Generate user-facing explanations for recommendations (XAI). Address agent risks: memory poisoning, goal manipulation, identity spoofing. Compare frameworks (LangChain, LlamaIndex, CrewAI, AutoGen) to match goals.

Apply prompt and project structuring best practices for agentic coding. Real-World Application & Use CasesWe design a proactive companion that monitors interests, fetches fresh articles, updates a preference model from likes/dislikes, and iterates autonomously—illustrating agent planning, tool use, and memory in a compact workflow. You’ll see how to evolve from a simple loop to a production-style recommender: topic extraction, weighted exploration vs.

exploitation, semantic vectors, source allow-listing, recency decay, and a user-readable “why this was recommended” message. Course Format & Learning ExperienceStructured modules combine concept briefings with hands-on labs: set up the environment and MCP scaffolding; implement RAG; wire Tavily+ChatGPT; add persistence (SQLite → ChromaDB); introduce topic pillars & semantic re-ranking; add explanation UX; then harden with tests and risk mitigations. Expect checklists, prompts, and refactors aligned to engineering best practices.

InstructorTaught by Shreejit Gangadharan with 12 years of industry experience in companies like Flipkart, Microsoft, and Google. Updated with 2024–2025 practices across MCP orchestration, vector stores, topic-pillar ranking, source allow-listing, recency decay, and agent risk controls. Primary Topics/Keywords: Agentic AI, Perceive-Reason-Act, MCP workflow, RAG, LangChain, LlamaIndex, CrewAI, AutoGen, Tavily, ChatGPT/Gemini, SQLite, ChromaDB, embeddings, semantic re-ranking, topic pillars, explainability, agent risks.

Prerequisites:Working knowledge of Python and virtual environments; comfort with CLI & Git. Understanding of HTTP APIs and JSON; basic familiarity with LLM prompts. Tools & Frameworks Used: LangChain, Tavily, ChatGPT/Gemini, SQLite, ChromaDB, pytest.

Capstone Project: CLI “Personalized News Curator” with preference learning, topic pillars, and semantic re-ranking, plus user-facing explanations.