1-Day Immersive Agentic AI Bootcamp
A fast-paced, high-intensity 1-day developer bootcamp focused entirely on building, orchestrating, and deploying autonomous AI agents. Go from basic API scripting to multi-agent stateful orchestration.
Program Objectives
By the end of this intensive day, participants will be able to:
- Understand the paradigm shift from prompt engineering to agentic workflows.
- Design and implement autonomous agent control loops using the ReAct (Reason + Act) pattern.
- Expose native Python functions and third-party APIs (e.g. search, computation) as LLM tools.
- Orchestrate complex agentic behavior with state management and cyclic loops using LangGraph.
- Implement multi-agent team patterns and utilize the LLM Council/Debate pattern.
- Deploy working interactive agent prototypes onto Streamlit and Hugging Face.
Target Audience
This bootcamp is tailor-made for software engineers, AI developers, researchers, and technical architects who already possess programming experience and want to rapidly master practical agent construction.
Prerequisites
Strong proficiency in Python (async/await, type hints, basic OOP) is required. A laptop with local Python 3.10+ installed and a valid API key setup (instructions below) are mandatory.
Bootcamp Schedule
| Slot | Time | Duration |
|---|---|---|
| Session 1: Foundations & ReAct Pattern | 9:30 AM – 11:00 AM | 90 min |
| Morning Coffee Break | 11:00 AM – 11:15 AM | 15 min |
| Session 2: Tool Integration & Research Agents | 11:15 AM – 1:00 PM | 105 min |
| Networking Lunch | 1:00 PM – 2:00 PM | 60 min |
| Session 3: Orchestrating Graphs with LangGraph | 2:00 PM – 3:30 PM | 90 min |
| Afternoon Refreshment Break | 3:30 PM – 3:45 PM | 15 min |
| Session 4: Multi-Agent Systems & Deployment | 3:45 PM – 5:30 PM | 105 min |
Pre-Bootcamp Setup Guide
Since this is an intensive, single-day bootcamp, participants must complete these setups before arrival. There will be no time allocated for environment troubleshooting during sessions.
1. Accounts & API Keys (All Free)
| Account | Purpose | Link |
|---|---|---|
| Google Account / AI Studio | To generate a free Gemini API key | aistudio.google.com |
| Hugging Face Account | Accessing open source models & HF Spaces | huggingface.co |
| GitHub | Hosting code and deployment pipelines | github.com |
| Streamlit Cloud | Free interactive interface hosting | streamlit.io/cloud |
2. Local Environment Installation
- Python 3.10+ — Install Python from python.org
-
Ollama (Local Models) — Install and download a lightweight model for offline testing:
ollama pull llama3.2:3b -
Required Python Libraries — Execute the following command in your virtual environment:
pip install langchain langchain-google-genai langgraph crewai streamlit duckduckgo-search wikipedia arxiv python-dotenv
Detailed Session Breakdown
Session 1 — Agentic AI Foundations & ReAct Pattern
Topics: Autonomy spectrum, loops, scratchpads, prompt-based ReAct engine.
Concepts & Theory
- The Spectrum of AI: Chatbots vs. Chains vs. Autonomous Agents.
- How LLMs plan and reason: Scratchpads, chain-of-thought, and self-correction.
- The ReAct loop pattern (Reasoning + Acting) explained mathematically and visually.
- Token limits, context window conservation, and state persistence.
Hands-on Lab
- Building a ReAct Agent from scratch: Writing a bare-metal Python agent loop that queries a Google Gemini model, processes thoughts, calls mock arithmetic tools, and parses final answers without frameworks.
Session 2 — Native Tool Integration & Web-Search Agents
Topics: Function calling, API schemas, Pydantic, search and extraction pipelines.
Concepts & Theory
- How function calling works under the hood: JSON schema generation, model guidance, and output intercepting.
- Type enforcement and schema definitions using Pydantic.
- Handling tool execution exceptions: retry limits, model corrections, and fallback values.
Hands-on Lab
- Building a Smart Research Agent: Equipping our agent with DuckDuckGo Search, Wikipedia API, and arXiv API. Constructing a system that accepts a complex topic, queries across multiple databases, handles scraping errors, and synthesizes a structured Markdown report.
Session 3 — Orchestrating Stateful Graphs with LangGraph
Topics: Node-edge graphs, state schemas, conditional routing, cyclic execution, persistence.
Concepts & Theory
- Why linear agent chains fail for complex, cyclic business processes.
- The state-machine abstraction: Nodes (functions), Edges (routing), and State (context database).
- Conditional execution, cycle checks, and human-in-the-loop validation checkpoints.
Hands-on Lab
- Building an Editorial Multi-Step Writer: Implementing a graph-based LangGraph application with a "Writer Node" and a "Critic Node". The agent generates an article, routes it to the critic for evaluation, cycles back for revisions if edits are suggested, and saves the final output only when requirements are satisfied.
Session 4 — Multi-Agent Systems, Council Patterns & Deployment
Topics: CrewAI roles, supervisor pattern, LLM Council, Streamlit deployment, security.
Concepts & Theory
- Multi-agent topologies: Sequential pipelines, hierarchical committees, and peer-to-peer discussions.
- The LLM Council Pattern: Orchestrating multiple distinct models (cloud Gemini vs. local Llama via Ollama) to debate, vote on, and optimize outputs.
- Security and safety: Preventing prompt injection, setting execution sandboxes, and rate limits.
Hands-on Lab
- Deploying the Agentic Application: Wrap the research agent in a web-based **Streamlit** user interface. Add configuration options for models, and deploy the live app to Streamlit Community Cloud.
Assessment & Certification
Hands-on Code Evaluation
During Session 4, participants showcase their working Streamlit/LangGraph codebase locally. Successful invocation of the tools, state checks, and output correctness are assessed live.
Certificate of Mastery
A digital Certificate of Mastery in Agentic AI Engineering (signed by the training organizer and referencing the syllabus design by Akshar Prabhu Desai) is awarded upon demonstrating a functioning agent workflow.