Generative AI Bootcamp: LLM Engineering, RAG & AI Agents

Learn how base, chat, and reasoning models differ in training and behavior, from next-word prediction to chat persona and chain-of-thought reasoning, and how to choose them by task and latency.

Explore how the transformer architecture powers GPT and modern llms, with self-attention enabling parallel processing. Understand context windows, token costs, and the evolution from GPT-1 to GPT-4, including RLHF.

Understand the physics of AI parameters, brute-force scaling laws, and open-source approaches like LLaMA and DeepSeek, linking model size to hardware needs and data quality.

Explore how text becomes tokens for large language models, comparing character-, word-, and token-level approaches, and explain why tokenization drives cost, context length, and pricing.

Examine how byte pair encoding tokenizes text into tokens, creating vocabulary limits. Understand math and spelling blind spots, multilingual cost differences, and why engineers split tasks between LLMs and code.

Learn to count tokens with the tick token library in Python to predict AI costs, and explore how tokenization, memory illusion, and statelessness shape LLM architecture and conversation state management.

Explore production constraints of LLMs—context window, output limits, and API billing—and apply RAG, prompt caching, and tiered routing to build reliable AI applications.

Identify the most strategic pages from a company website with a rich system prompt, then output a structured JSON payload of category and absolute URL for the corporate brochure.

Master prompt chaining to orchestrate multiple llms in sequence, build site knowledge from scraped content, and synthesize it into a professional brochure.

Demonstrates end-to-end execution of an AI brochure pipeline, from compiling site intelligence and JSON extraction to streaming and rendering a corporate brochure in real time.

Explore building the Frontier Arena multi-cloud, multi-model architecture with Open AI, DeepSeek, and Grok XAI to achieve API portability, vendor-agnostic LLM pipelines, and seamless model swapping.

The paradigm shift moves from fast, impulsive models to reasoning models that scale compute for planning and verification, highlighting system 1 and system 2 thinking and GPT-5.

Benchmark frontier AI models with a universal connector pattern to compare accuracy, latency, and cost across GPT-5, DeepSeek, and Grok in a multi-cloud arena.

Implement local inference with ollama and open router to decentralize AI, run models on your laptop with zero API cost, and use open router as a failover bridge.

Examine an abstraction layer that tames SDK chaos by comparing LangChain and the Lite LLM for production, highlighting lightweight microservice tools versus heavyweight frameworks to reduce technical debt.

Master rapid prototyping of AI apps with Gradio in Python, no HTML, CSS, or JavaScript required. Leverage gr.interface, gr.chat, and gr.blocks to handle UI, routing, and memory for user feedback.

Explore gradio state and callbacks to create memory across prompts, enabling stateful interactions. Learn how to deploy a public URL and manage memory and events with gr.state.

Secure the engine by implementing built-in gradio authentication to lock the ai interface. Wire the secure ui directly to the llm backend and support multimodal model selection.

Enable real-time streaming in the front-end ux by piping ai tokens via a python generator into gradle ui to render markdown and code blocks as they arrive.

Merge the back-end arena with a streaming front-end to run a real-time, side-by-side multi-model dashboard comparing OpenAI and DeepSeek on the same prompt.

Master the LLM context window as RAM, using zero-shot and multi-shot prompts to shape output and persona. Combat hallucinations with RAG, retrieving private data via vector embeddings for meaning-based search.

The lecture explains how text embeddings convert language into 1536-dimensional vectors to enable semantic search, not lexical matching, and use rag with cosine similarity to retrieve relevant paragraphs.

Load diverse documents using standard loaders to sanitize data for embedding models. Design chunking strategies with fixed size and overlap to preserve context for precise semantic search in RAG systems.

Build the final rag pipeline that grounds ai in private data by ingesting, chunking, embedding into a vector database, retrieving context with cosine similarity, and streaming grounded answers.

Demystify tool calling and show how prompt engineering with json tool schemas gives the ai hands to call tools and perform live actions, not autonomous execution.

Explore the agentic paradigm, distinguishing retrieval-augmented generation from live-action tools, and apply a decision-tree to choose rag or tools for live data, state-changing tasks, and CRM updates.

Demonstrates building airline assistant with a single tool execution, binding a Python function getFlightStatus to an llm via a json schema, and implementing an orchestration loop for autonomous tool use.

Learn to run flight and weather checks in parallel with a Python backend and thread pools, leveraging OpenAI's parallel tool calls to cut latency.

Wire the LLM into a local SQLite database to translate natural language into SQL, execute queries via a Python backend, and summarize findings.

Build an autonomous agent loop that thinks, acts, observes, and orchestrates multiple tools to keep users satisfied, applying the agentic loop pattern for tool-dependent tasks.

Elevate the interface layer to a multimodal canvas that handles vision, voice, and generation, using Gradio to map data types for images and audio and manage heavy binary data.

Build an end-to-end multi-modal agent that fuses vision and voice by integrating DALL-E 3 image generation and text-to-speech, routing prompts to text, image, or audio outputs in a Gradio interface.

Deploy an enterprise omni agent by wiring a multimodal loop into a secure Gradio dashboard, using LLMs, SQL execution, and chart image generation.

Master the basics of Python syntax and semantics, including single and multi-line comments, the print statement, and how indentation and case sensitivity govern code structure.

Master Python’s dynamic typing, type inference, and runtime variable updates, while recognizing common syntax errors, case sensitivity, indentation, and f-string formatted prints in conditional blocks.

Explore Python operators, including arithmetic, comparison, assignment, identity, membership, and ternary operations, with practical examples in a Jupyter notebook to boost program efficiency.