From Generative AI and LLM engineering to RAG, LangChain, LangGraph and multi-agent workflows. Learn by building practical AI applications that retrieve knowledge, use tools and take controlled actions.
This course teaches you to build applications that use documents as knowledge, create dependable LCEL pipelines, serve applications through APIs, and coordinate agents through structured workflows. Learn approaches across LangGraph, Google ADK and OpenAI Agents SDK, with optional fine-tuning and MCP modules.
Eleven modules, progressing from fundamentals to practical multi-agent systems.
Python classes and objects; Jupyter notebook workflow; basic deep-learning examples, data preparation and model intuition; Transformer, Mixture-of-Experts and multimodal model concepts.
Type-safe data with BaseModel and Python type hints; optional values, lists, nested models and validation rules; structured data for APIs and LLM outputs.
Load text, PDF, XML and other document sources; recursive, character, HTML and JSON splitters; embeddings and knowledge retrieval with FAISS and Chroma.
Fast inference with ChatGroq; LCEL pipelines: prompt | model | parser; chatbot history, service and client workflows.
Model integration and message-based conversations; custom tools and tool-calling loops; structured output and middleware controls.
Graph state, nodes and edges; simple graphs and chatbot conversation flows; how state moves through a multi-step process.
Chains as reusable graph components; tools and ReAct-style agents; routing decisions and streaming responses.
Prompt chaining for focused stages; parallelization for independent work; routing to the right path based on input or results.
Compare RAG architectures and vectorless approaches; guardrails for LLM inputs and outputs; orchestrate retrieval and response with LangGraph.
Build the same two-agent sales pipeline three times: a Sales Analyst checks data against targets and a Report Agent creates a daily report. Compare LangGraph StateGraph, Google ADK SequentialAgent, and OpenAI Agents SDK, all using Groq llama-3.1-8b-instant.
LoRA: DistilBERT adapter fine-tuning for sales review classification. QLoRA: Qwen2.5-0.5B 4-bit fine-tuning (NVIDIA CUDA GPU required). MCP: expose sales_by_region and top_product Python functions as discoverable tools with FastMCP.
Load AI research papers from a folder, split and embed them in Chroma, retrieve relevant passages, and answer grounded questions from a local Streamlit interface.
Specialised agents analyze customer conversations and combine findings into business recommendations.
Ask questions about sales data in plain English; generate and execute Matplotlib chart code, produce findings and recommendations, and optionally email the report.
Agent 1 writes a safe MySQL SELECT from the supplied schema; Agent 2 runs it and returns a downloadable table.
Basic Python and a willingness to experiment. Set up Python, VS Code and Jupyter. Provider API keys may be needed for Groq, OpenAI or Gemini. Ollama supports local models; MySQL is used in Project C, and an NVIDIA CUDA GPU is required for the QLoRA add-on.
Curriculum adapted from DhishaAI’s 8-page Agentic AI Student Overview. Optional modules are identified above.
New batch starts 24 October 2026. Enquire about the course and registration.
Enquire Now ↗