Arun Prakash Asokan

Agentic RAG adds a “brain” to the RAG pipeline – bringing reasoning, tool use, and adaptiveness – which translates to tangible business value in accuracy, flexibility, and user trust

This workshop is a deep dive into Agentic RAG (Retrieval-Augmented Generation) – an emerging approach that combines the power of LLM-based agents with retrieval techniques to build smarter AI applications. Over an 8-hour session (of course including breaks in between), participants will explore how to move beyond “vanilla” RAG pipelines and infuse them with agentic behavior for greater flexibility and intelligence.

The full-day (8-hour) session employs Google Colab notebooks for immersive practice. Participants will work with:

• Azure Document Intelligence for complex document parsing and handling multi-modal content (text + images).
• Chroma Vector DB for semantic storage and retrieval.
• LangChain for application development.
• LangGraph for agent orchestration and complex workflow management.
• Langfuse for agentic tracing, observability, and evaluation.

Big Business Benefits of Agentic RAG

Agentic RAG empowers large-scale, enterprise-ready AI systems by combining retrieval-augmented generation with intelligent, decision-making agents. The result? Smarter, more reliable, and adaptable GenAI solutions.

• Higher Accuracy & Trust – Agents reason, validate, and reduce hallucinations.
• Multi-Modal Intelligence – Combine text, image, and structured data seamlessly.
• Observability & Evaluation – Langfuse for full agent tracing and performance insights.
• Dynamic Multi-Source Retrieval – Use Chroma vector DB + Travily web search + API & DB integrations.
• Autonomous Problem Solving – Complex queries handled via reasoning + tool orchestration.

💡 Think of Agentic RAG as adding a decision-making brain to your RAG pipeline—boosting precision, flexibility, and business value.

Workshop Flow

We will start with the Capstone Project overview—a real-world, multi-modal Agentic RAG application—explaining all the complex concepts we will learn and implement during the day. This includes:

• Multi-source data ingestion (files, web, APIs, images).
• Complex document parsing with Azure Document Intelligence.
• Multi-modal retrieval and reasoning.
• Agent orchestration with LangGraph.
• Observability & evaluation with Langfuse.

Prerequisite: 

• Basic Python skills 
• Introductory level exposure to LLMs and RAG
• Familiarity with LangChain or similar frameworks 
• Exposure to vector databases and embeddings

*Note: These are tentative details and are subject to change.

Retrieval-Augmented Generation (RAG) has been a game-changer for grounding LLMs with relevant context. But in real-world scenarios, traditional RAG and sometimes Agentic RAG can hit limitations—especially when reasoning, relationships, and domain context become more complex.

This is where Agentic Knowledge Augmented Generation (Agentic KAG) comes in. By combining Knowledge Graphs, Graph Databases, and AI Agents, we enable LLMs to generate knowledge-rich, contextual, and deeply insightful outputs—all with traceability, transparency, and reasoning baked in.

In this live hack session, we will:

• Build a Knowledge Graph from scratch using an interesting, relatable dataset that every DataHack Summit 2025 attendee will understand. Yes… it is the data about the DHS events itself. 🙂
• Use Neo4j Graph Database as the backend to power graph traversal and reasoning.
• Use Chroma DB for Traditional RAG and Agentic RAG scenarios.
• Explore ReAct Agents for step-by-step reasoning and tool usage. Build and end2end pipeline for Agentic KAG. 

Implement Langfuse for Agentic traceability and observability—so you can see exactly how your agents think, retrieve, and decide.