Online Algorithmic Trading

The course begins with an overview of what LLMs are—massive neural networks trained on vast amounts of text data capable of understanding and generating human-like language. We'll explore the high-level architecture and working principles behind these models, including the role of transformers, attention mechanisms, and pretraining.

The course dives into the different types of LLMs and how to access them via APIs. We’ll cover widely used models like OpenAI's GPT, Anthropic’s Claude, Meta’s LLaMA 3, and DeepSeek, discussing their strengths and common use cases. We’ll also explore how to get API access, authenticate, and integrate these models into applications.

A special module will introduce Ollama, a powerful tool for running LLMs locally, and show how to work with Grok APIs for real-time data integration. Learners will gain insight into tokenization and prompt engineering basics, including how to craft effective prompts, use few-shot and zero-shot techniques, and structure inputs for optimal outputs.

We’ll also contrast fine-tuning vs. prompt engineering—explaining when to use each approach for customizing model behavior. To help learners design efficient AI apps, we’ll cover key performance considerations, such as model latency, cost optimization, and response quality.

The topic concludes with a hands-on project, where participants will build a chatbot using the OpenAI API—tying together all the concepts into a practical, real-world application. This curriculum is ideal for developers and AI enthusiasts looking to break into the world of LLMs and build intelligent, responsive AI solutions.

This module offers a hands-on, structured introduction to LangChain, a powerful framework for building applications using large language models. We begin with the basics of LangChain, exploring its purpose, benefits, and why it's emerging as a go-to tool for building LLM-powered applications. Participants will learn about LangChain’s architecture and core components, followed by a practical walkthrough on setting up the development environment using popular IDEs such as Cursor AI, Jupyter Notebook, and VS Code. We’ll configure APIs from providers like OpenAI, DeepSeek, and Grok, and introduce Ollama for local LLM deployment. Learners will gain a solid grasp of LLM integration with LangChain, including prompt engineering concepts—covering prompt structures, templates, zero-shot and few-shot prompting, chaining, parsing outputs, and saving/loading prompts.

We’ll explore document handling with document loaders, text splitters, and indexing. A deep dive into LangChain’s chain types—including sequential chains, LLM router chains, transform chains, MathChain, and building custom chains—helps learners understand modular flow designs. Memory is another crucial component; we’ll cover memory management, types of memory, and how to implement them in conversational AI, including conversational buffer and entity memory.

The course concludes with advanced tools and integrations, teaching how to connect LangChain with external APIs, build backends using FastAPI, and integrate with SQL and vector databases for knowledge-driven LLM applications. This comprehensive program equips learners to build robust, scalable AI applications using LangChain’s modular ecosystem.

In this module, learners will dive deep into Vector Databases and their critical role in powering modern Retrieval-Augmented Generation (RAG) and Agentic RAG applications. We begin by understanding embeddings and similarity search, covering distance metrics such as cosine similarity, Euclidean distance, and k-nearest neighbors (k-NN). Learners will explore the landscape of open-source vector databases, including FAISS, ChromaDB, Qdrant, and Weaviate, and understand where and how each can be applied. Through hands-on sessions, we’ll cover indexing, embedding, and querying using ChromaDB, as well as integrating Qdrant with LangChain to build a basic RAG pipeline.

We’ll then transition into the core principles of RAG, covering its architecture, components, and tokenization strategies, and how vector search integrates into LLM pipelines. Learners will understand how RAG compares to finetuning and prompt engineering, and how chunking strategies (semantic vs. fixed-size) impact performance. This leads into advanced retrieval techniques such as metadata filtering and hybrid search, culminating in the development of an Agentic RAG application, where multiple agents collaborate to retrieve and reason over information.

Learners will also explore embedding-powered AI agent use cases, understand LLM integration via LangChain, and examine how enterprises use vector databases for personalization, intelligent search, and AI-driven recommendations.

The course wraps up with two real-world projects: a Document Q&A Bot using ChromaDB or Qdrant with OpenAI, and an AI Agentic RAG system. These capstone projects reinforce concepts around knowledge storage, retrieval, and building scalable, intelligent GenAI systems.

This module introduces learners to the world of AI Agents—autonomous, goal-driven entities that can perceive, reason, and act within an environment. We begin by understanding what AI agents are, and how they differ from traditional AI applications, which are typically static or rules-based. The course dives into the key components of an AI agent, including perception, memory, reasoning, and action layers, followed by a discussion on the different types of agents such as reactive agents, conversational agents, and planning agents.

We also explore the distinction between single-agent and multi-agent systems, and how these paradigms are applied in real-world use cases like research assistants, support bots, or autonomous systems. The hands-on part of the module focuses on integrating agents with external APIs, performing web scraping, and leveraging tools that give agents access to dynamic, real-time data—enabling them to make context-aware decisions. Learners will walk away with a clear understanding of how modern AI agents can be embedded into business and consumer applications, driving automation, personalization, and real-time intelligence.

This module provides an in-depth exploration of CrewAI, a powerful Python framework designed to build and manage multi-agent AI systems. We begin by understanding what CrewAI is and how it enables the orchestration of multiple agents that can collaborate on complex tasks. Learners are introduced to its core concepts—Agents, Tasks, Tools, and Crews—and learn how to define each agent's roles, goals, and capabilities within a system.

We cover advanced topics like role definition tuning using backstories and expectations, as well as task decomposition strategies, distinguishing between sequential and hierarchical workflows. The curriculum walks through how to create and manage multi-agent workflows, enabling agent collaboration for more intelligent and distributed problem-solving.

In the hands-on section, learners will build a Research Assistant using CrewAI, integrating LLMs and external APIs like Serper, Google, and Bing for real-time data retrieval. By the end of the module, participants will be equipped to develop scalable multi-agent applications that mimic real-world human collaboration with AI agents working in coordinated crews.

Model Context Protocol (MCP) is a conceptual framework or standardized structure that governs how context is managed, shared, and updated between large language models (LLMs), agents, tools, and tasks in an AI system—particularly in multi-agent and agentic AI environments. MCP ensures that agents can collaborate effectively, share relevant data, and make context-aware decisions. This is critical in building scalable, transparent, and efficient AI systems where agents need to work autonomously or in coordination with other agents and APIs.

In this module, learners will first be introduced to the need for contextual coherence in AI agents, followed by the principles of the Model Context Protocol. The course begins by examining the importance of context in LLM-based workflows, particularly in multi-agent environments like CrewAI or LangGraph. We will explore how MCP supports persistent memory, task execution flow, and inter-agent collaboration. Next, the curriculum dives into the key components of MCP, including agent identity, shared memory, prompt inputs/outputs, task metadata, tool integration logs, and error handling. The hands-on section will guide learners in building an agent framework where tasks are context-aware, with state passed via structured dictionaries or JSONs. Learners will use CrewAI or LangGraph to simulate multi-agent workflows, implement memory buffers, and design a basic MCP layer that orchestrates interactions between retrievers, reasoners, and responders in an agentic RAG system. The module concludes with best practices in debugging and optimizing model latency, as well as strategies to store and retrieve structured memory for long-term applications.

Detailed explanation of the Capstone Projects and Case Studies module for your AI Agent development training program:

Capstone Projects & Case Studies

This final module is designed to consolidate all learning through real-world, hands-on projects that blend theory, coding, system integration, and creative AI design. Learners will apply their knowledge of LangChain, Vector Databases, LLMs, and Multi-Agent frameworks like CrewAI to build end-to-end AI applications.

1. Voice-Based AI Agent

Objective: Build a conversational agent that users can interact with using voice input and receive natural language responses. Key Features:

• Speech-to-Text integration (e.g., using Whisper or Google Speech API)

• LangChain-powered LLM backend for response generation

• Text-to-Speech output for seamless voice feedback

• Contextual memory management using LangChain's buffer or entity memory Use Cases:

• Voice-based virtual assistants for websites

• Customer support bots

• Accessibility tools

2. Agentic RAG (Retrieval-Augmented Generation with Agents)

Objective: Design an advanced RAG pipeline where agents autonomously retrieve, reason, and respond using a vector database and LLM. Key Features:

• Integration with ChromaDB or Qdrant for document storage and semantic search

• Intelligent agents to retrieve relevant information, summarize, and answer

• Metadata filtering and hybrid search to improve retrieval precision

• Use of LangChain or CrewAI for managing agent workflows Use Cases:

• Enterprise knowledge assistants

• Internal document search tools

• Legal or policy Q&A bots

3. Multi-Agent AI Chatbot (Built with CrewAI)

Objective: Build a multi-role conversational system where each agent has a specific capability and collaborates to serve the user. Key Features:

• Multiple CrewAI agents (e.g., Researcher, Summarizer, Analyst, Presenter)

• External API integration (Serper for web search, SQL for data, etc.)

• Goal-based task delegation using CrewAI's orchestration

• Fully deployed with FastAPI backend and Streamlit frontend Use Cases:

• Executive assistant bot

• Market research automation

• Interactive tutor or coach

Outcomes & Deliverables

• Learners submit a fully functional AI application (code + demo video)

• Project report explaining architecture, tools used, and challenges faced

• Peer-reviewed or instructor-evaluated feedback

• Certificate of Completion with Project Highlight