Trustworthy RAG & Autonomous AI Agents: Build Retrieval, Reasoning & Tool-Using Systems

This 3-day virtual hands-on workshop introduces participants to Retrieval-Augmented Generation (RAG), autonomous AI agents, and trustworthy AI systems. Participants will learn to connect LLMs with external knowledge, vector databases, tools, and APIs for grounded and reliable outputs. The workshop covers semantic and hybrid retrieval, agent memory, tool calling, reasoning, workflow orchestration, and RAG evaluation. It also addresses hallucination mitigation, guardrails, observability, and human-in-the-loop validation. Through guided hands-on sessions, participants will build and evaluate a practical RAG-powered autonomous AI assistant.

The aim of this workshop is to equip participants with practical skills to build grounded RAG systems and autonomous AI agents . It also focuses on improving reliability through evaluation, hallucination mitigation, guardrails, and trustworthy AI practices.

• Understand the fundamentals ofRetrieval-Augmented Generation, LLM grounding, and knowledge retrieval.
• Build RAG pipelines using embeddings, vector databases, semantic search, and hybrid retrieval.
• Developautonomous AI agentswith tool calling, memory, reasoning, and workflow orchestration.
• Evaluate AI systems for factuality, faithfulness, hallucinations, and retrieval quality.
• Applytrustworthy AI practices, including guardrails, observability, and human-in-the-loop validation.

📅 Day 1: Retrieval-Augmented Generation & Knowledge Grounding

• Introduction to Large Language Models and knowledge limitations Hallucinations, outdated knowledge, and domain-specific information gaps Fundamentals of Retrieval-Augmented Generation Document ingestion, chunking, embeddings, and indexing Semantic search, keyword retrieval, and hybrid retrieval Vector databases and similarity search Metadata filtering and reranking Prompt construction with retrieved context Citation-aware and source-grounded responses Introduction to knowledge graphs and advanced RAG architectures
• 🛠️ Hands-on:
• Load and preprocess domain-specific documents Generate embeddings and create a searchable knowledge base Build a basic RAG question-answering system Compare keyword search, semantic retrieval, and hybrid retrieval Generate grounded answers with supporting context and citations
• 🧰 Tools Covered:
• Google Colab, Python, LangChain/LlamaIndex, Hugging Face, Sentence Transformers, FAISS/Chroma, BM25, and LLM APIs

📅 Day 2: Autonomous AI Agents, Tool Use & Reasoning

• From conversational AI to autonomous AI agents Core components of an agent: model, tools, memory, state, and planning Agentic RAG and retrieval as an AI-agent capability Tool/function calling and structured outputs Reasoning, task decomposition, and multi-step workflows ReAct-style reasoning and action cycles Short-term and persistent agent memory Agent routing and workflow orchestration Self-reflection, verification, and adaptive decision making Single-agent and multi-agent architectures Human-in-the-loop AI systems Applications in research, enterprise automation, analytics, and scientific problem solving
• 🛠️ Hands-on:
• Convert the Day 1 RAG pipeline into a tool-using AI agent Add document search, calculation, and structured-data tools Implement tool selection and task routing Build a multi-step reasoning workflow Add memory and a verification/reflection step Develop a prototype autonomous research or enterprise assistant
• 🧰 Tools Covered:
• Google Colab, Python, LangGraph, LangChain/LlamaIndex, LLM function calling, Pydantic-style structured outputs, FAISS/Chroma, and APIs

📅 Day 3: Trustworthy AI, Evaluation & Agent Reliability

• Principles of trustworthy and responsible AI Factuality, faithfulness, grounding, and answer relevance Hallucination detection and mitigation Retrieval-quality and generation-quality evaluation Context precision, context recall, and answer correctness Uncertainty, confidence, and calibration Prompt injection and adversarial RAG Tool misuse and unsafe agent behaviour Bias, privacy, fairness, and responsible AI Explainability and traceable agent workflows Human-in-the-loop evaluation Guardrails, auditing, and observability Evaluation benchmarks for RAG and agents Trustworthy deployment in research, healthcare, finance, education, and enterprise systems
• 🛠️ Hands-on:
• Build an evaluation dataset for the RAG-agent system Compare baseline LLM, RAG, and agentic-RAG responses Evaluate retrieval relevance, groundedness, and faithfulness Detect unsupported or hallucinated responses Add verification, reranking, and guardrails Trace agent actions and tool calls Prepare a final trustworthy AI evaluation report
• 🧰 Tools Covered:
• Google Colab, Python, RAGAS/DeepEval, LangGraph, LangSmith-style tracing, FAISS/Chroma, reranking models, pandas, and evaluation templates

• Understand the architecture and limitations of modern LLM-based systems.
• Build a completeRetrieval-Augmented Generation pipelineusing domain-specific documents.
• Implement semantic, keyword, and hybrid information retrieval.
• Create grounded AI responses with supporting evidence and citations.
• Transform a conventional RAG system into atool-using autonomous AI agent.
• Design multi-step reasoning and task-decomposition workflows.
• Integrate external tools, databases, and APIs with AI agents.

Graduate and Postgraduate Students in Computer Science, AI, Data Science, IT, Software Engineering, Electronics, and related fields PhD Scholars and Researchers working in LLMs, NLP, RAG, knowledge systems, autonomous agents, or trustworthy AI Academicians and Faculty Members interested in Generative AI, intelligent systems, research applications, and curriculum development Industry Professionals such as AI/ML engineers, data scientists, software developers, solution architects, and automation professionals Participants with basic Python knowledge are preferred; prior experience with RAG or AI agents is not mandatory