AI4SE Tutorial

Welcome to AI4SE Tutorial

Discover the power of Artificial Intelligence in Software Engineering with this interactive tutorial.

Lets set things up.

Follow this notebook for the Tutorial: Tutorial Notebook

Now we setup Gemini. Do the following:

• Generate Your API Key
  • Go to Google AiStudio Api Key
  • Generate Key.
  • Provide any necessary permissions when prompted.
• Add Your API Key to Colab
  • Open your Google Colab notebook.
  • On the left panel, click the 🔑 “Secrets” icon.
  • Click “+ Add a new secret”.
  • ‘Name’ is you key name, and ‘Value’ is the key
• Configure the API Key in Your Code
  • Change the name in the 3rd cell if needed: GOOGLE_API_KEY=userdata.get(‘Key_Name’)
  • Colab will ask for permission to access your secrets. Grant access to proceed.

---

Content

1. Prompting

Image credit: The Power of Prompt Engineering: Refining Human -AI Interaction with Large Language Models in The Field of Engineering

💡 Prompting involves crafting effective instructions or queries to guide AI models (LLMs) in generating desired outputs. 🧠 LLMs might often need a greater context (History) rather than just the user input to generate a response.

2. Function Calling

Image credit: Expanding AI Horizons: The Rise of Function Calling in LLMs

🔧 Function calling lets a language model go beyond text generation by triggering external tools or APIs based on the user’s prompt. When the model detects that a specific task—like retrieving weather data, running code, querying a database, or automating a process—is needed, it calls a predefined function with the appropriate inputs.

These functions can do things like:

• Access cloud services (e.g., weather or maps)
• Run backend processes or scripts
• Query databases
• Execute code or automation tasks

🤖 The model then uses the function’s output to generate a final, informed response. This turns the model into a powerful agent that can interact with external systems dynamically.

3. Retrieval-Augmented Generation (RAG)

Image credit: How I built a Simple Retrieval-Augmented Generation (RAG) Pipeline

1. Data Indexing

This stage prepares documents for retrieval later.

• Data Loading: Raw documents are collected and loaded.
• Data Splitting: Documents are split into smaller text chunks (e.g., paragraphs, sentences).
• Data Embedding: Each chunk is converted into a vector embedding using an embedding model. These are numerical representations that capture the semantic meaning of the text.
• Data Storing: The embeddings are stored in a Vector Database (Vector DB), which enables fast similarity search.

2. Data Retrieval & Generation

This stage is triggered when a user asks a question.

• User Query: A user submits a question.
• Vector Embedding: The query is converted into a vector (like the document chunks).
• Retrieval: The vector DB is queried with the query embedding to retrieve the most relevant Top-K Chunks.
• Generation:
  • The retrieved chunks are passed to a Large Language Model (LLM).
  • The LLM uses this context to generate a response.

4. Agentic AI

Image credit: What is Agentic AI | K21Academy

🤖 Agentic AI refers to AI systems that can act as autonomous agents, making decisions, using tools, collaborating with others, and adapting to complex tasks over time.

🔧 Agent Customization Agentic AI allows you to build specialized agents by customizing:

• Capabilities (e.g. access to specific tools or APIs)
• Behavior (how they interact or respond)
• Roles (like data analyst, assistant, or coder)

🧠 Agent Architecture An agent can be powered by

• Memory (to remember past actions)
• Tools (to interact with the world, like APIs or functions)
• Goals (what the agent is trying to achieve)

🤖 Multi-Agent Conversations Agentic systems involve multiple agents working together. This supports collaborative problem-solving, with agents sharing results or coordinating workflows. They can have various communication patterns.

• Joint chat: Multiple agents interact simultaneously.
• Hierarchical chat: One lead agent delegates tasks to others.

🔁 Flexible Interaction Patterns Agents aren’t limited to user prompts—they

• Interact with each other
• Trigger actions autonomously
• Update plans dynamically

Image credit: AutoGen

---

Instructors