1.       Overview

ACE conducted an AI and ML Workshop on 29th August 2026 for the college. The session was conducted by fourth-year students of the Computer Department, who served as speakers for the event. The workshop combined theory sessions on core AI/ML and RAG concepts with hands-on coding activities, followed by a team-building problem-statement competition and a closing fun quiz round with prize distribution to the speakers.

2. Topics Covered

2.1 Introduction to RAG (Retrieval-Augmented Generation)

• RAG gives an LLM access to relevant external data before it generates an answer.
  • RAG helps in avoiding hallucination.

2.2 RAG Pipeline

The data preparation stage was covered first: before any retrieval can happen, raw documents are processed into a form the system can search efficiently. This involves:

• Document loading
  • Text extraction
  • Chunking (an optimization part of RAG)

2.3 Understanding the RAG Vector System

• Text is converted into numbers using transformers.
  • Semantic clustering
  • High dimensionality

2.4 Vector Embedding and Storing

After embedding, the data is stored in an embedded vector database for fast retrieval.

• Indexing — for optimization
  • Metadata — for contextual learning

2.5 Query Processing

The user’s query is converted into a query embedding, which is then checked for similarity against the vector embeddings in the database, using:

• Cosine similarity
  • Natural language processing (NLP)

2.6 Retrieval Techniques

• FAISS vector index — used to search a query against embedded data.
  • Keyword index (BM25-based).
  • Hybrid retrieval, combining vector and keyword search.
  • Query router — decides whether to use RAG, Gemini, Google Search, or a combination of these.

2.7 Generative AI and Transformers

• Overview of Generative AI (Gen AI).
  • Transformers were explained using ChatGPT as an example — a Generative Pretrained Transformer that predicts the next token after a question is asked.
  • Reference to the “Attention Is All You Need” research paper.
  • Neural networks and artificial neural networks (ANN).
  • Encoder layer, attention mechanism, and decoding layer.

2.8 Agentic AI

• Introduction to Agentic AI.
  • Introduction to Multi-Agent AI.

3. Hands-on Activities

Alongside the theory, students worked through hands-on notebooks covering machine learning, RAG-based application building, and agentic AI.

3.1 Machine Learning Hands-on

Students worked on a machine learning notebook using the Student_Performance.csv dataset, covering:

• Loading and preparing the dataset with pandas and NumPy.
  • Building a Decision Tree Regressor model.
  • Building a Linear Regression model with Polynomial Features to model a nonlinear relationship.
  • Evaluating models using metrics such as Mean Squared Error and R² score.
  • Feature importance analysis and visualization of results using Matplotlib and Seaborn.

3.2 RAG-based Study Assistant Hands-on

Students built a RAG-based Study Assistant application, covering:

• Installing and using libraries such as google-generativeai, sentence-transformers, faiss-cpu, rank_bm25, pypdf, python-docx, python-pptx, reportlab, and huggingface_hub.
  • Extracting text from PDF, DOCX, and PPTX study material.
  • Building a Hybrid Retriever combining FAISS (vector search) and BM25 (keyword search).
  • Integrating a Gemini-based generation engine to answer questions using retrieved context.
  • Building a Study Assistant that could summarize content, generate MCQs, and create flashcards from uploaded material.
  • Exporting generated content to PDF.
  • Generating educational images to accompany study material.
  • Uploading files to build a knowledge base and interacting with it through an interactive chat interface.

3.3 Agentic AI Hands-on

Students worked on an agentic AI notebook, covering:

• Using LangChain with a Gemini model (ChatGoogleGenerativeAI) to invoke an LLM.
  • Using DuckDuckGo Search as a tool to fetch live information (e.g., latest stock news).
  • Using the Tavily search API as an alternative search tool.
  • Defining custom tools such as a search tool and a get-current-time tool.
  • Creating an agent with a system prompt, instructing it to act as a stock research assistant that searches for information, uses tools, explains its reasoning, and avoids hallucination.
  • Running the agent to analyze stocks (for example, Tata Motors and Vedanta), covering latest news, earnings performance, growth prospects, industry trends, risks, and a final recommendation.

4. Team-Building Competition

At the end of the session, a team-building competition was held. Teams were required to select a problem statement (PS) and build a model to solve it, applying the concepts covered during the workshop.

5. Fun Quiz Round and Prize Distribution

The session concluded with a fun quiz round, followed by a prize distribution ceremony held for the speakers.