Introduction to AI applications

Intro to AI modalities

Examples of AI applications

Technologies

How to brainstorm AI applications

Vertical applications and their example technologies

Indie hacker examples

Venture funded LLM examples

This module introduces foundational concepts and practical workflows for working with Large Language Models (LLMs). Topics include terminology (e.g., ChatGPT vs. LLM, inference phases, training stages, and model compression techniques), the LLM ecosystem (vector databases, inference APIs, and fine-tuning libraries), and the model lifecycle. Participants will build a simple LLM-based system from scratch, starting with “Hello World” inference using Hugging Face, and deploy an LLM API using Modal for serverless deployment.

What is a tokenizer? What makes a tokenizer "good"?

Build a baseline 'word-based' tokenizer

Build a byte-pair encoding

Compare with state-of-the-art tokenizers (e.g., Llama’s)

What is an embedding? What makes an embedding "good"?

Understand the semantic meaning of word embeddings

Train a word embedding on Shakespearean tokens

Building a Shakespearean Language Model

What is an n-gram?

Train a 2-gram model

Predict using 2-gram model

Train an n-gram model

Predict using n-gram model, in an autoregressive manner

Building an n-gram language model

A minimal version of self-attention

Build a batched version of self-attention

Building self-attention

A minimal version of the feed-forward neural network (e.g., MLP)

Build a batched version of the MLP

“Train” on sanity check data

Building the feed-forward neural network

We will ignore multi-head attention as it introduces unnecessary complexity into the code. We can mention this in passing.

Add the remaining components (Skip connections, norms, positional encodings)

We will check that training loss decreases, but we won’t train until convergence.

Use black box optimizer to pretrain

Predict using transformer-based language model, autoregressively

Assembling the transformer-based language model

Run unsupervised evaluation using perplexity

Export the model using torchscript

Plug in our custom model into our LLM API

Open-sourced pre-trained models include Llama, Phi-3, Mixtral etc.

Evaluating and deploying a transformer-based language model

Pre-curated pretraining datasets

Common tools for data curation (mech turks, vis tools)

how to add guardrails

Datasets

Build a minimal version for our Shakespearean model

Build low-rank adapters for Shakespearean model

Vast repository of datasets for instruction tuning

Download instruction tuning dataset

Use huggingface’s peft on an actual pre-trained model, like Llama

Run low-rank adapter fine-tuning

Download and run inference. Compare with pre-trained model’s ability to follow instructions from a user.

Open-sourced instruction-tuned models include DBRX, Pythia, Cerebrus etc.

Low-Rank Adapters for Instruction Tuning

What is RAG?

Implement RAG

Apply RAG to Shakespearean model

Apply RAG to pre-trained Large Language Models

Integrate RAG into our API

Retrieval-Augmented Generation (RAG)

How to use LlamaIndex to chain tools together

The data wall

The ultimate test of time (Use fundamentals to understand and use time to filter)

How to keep to date

Multi modal large language models

The Future of Large Language Models

project scoping

data needs

modeling strategies

deployment requirements

Machine learning operations

What are agents?

Design patterns

Agents

In this bootcamp, we dive deep into Large Language Models (LLMs) to help you understand, build, and optimize their architecture for real-world applications. LLMs are revolutionizing industries—from customer support to content creation—but understanding how these models work and optimizing them for specific tasks presents unique challenges.

Over an intensive, multi-week curriculum, we cover:

The technical foundations of LLMs, including autoregressive decoding, positional encoding, and multi-head attention.
The LLM lifecycle—from large-scale pretraining to fine-tuning and instruction tuning for niche applications.
Industry best practices for model evaluation, identifying performance bottlenecks, and employing cutting-edge architectures to balance efficiency and scalability.
This bootcamp includes hours of in-depth instruction, hands-on coding sessions, and access to a dedicated community for ongoing support and discussions. Additionally, you’ll have exclusive access to code templates, an expansive reference library, and downloadable resources for continuous learning.

Your expert guides through this bootcamp are:

[Alvin Wan](https://www.linkedin.com/in/alvinwan/): Alvin is a Senior Research Scientist at OpenAI with over a decade of experience in AI, specializing in large language models and efficient AI design. Previously, he was a Senior Research Scientist at Apple, working on AI and large language models for Apple Intelligence. Alvin also worked on Tesla’s AutoPilot and holds a PhD from UC Berkeley, where his research has garnered over 3,000 citations. He brings a unique combination of industry expertise and cutting-edge research to this course, guiding you through the technical aspects of building, optimizing, and deploying LLMs.

[Zao Yang](https://www.linkedin.com/in/zaoyang/): Zao is a co-founder of Newline, a platform used by 150k professionals from companies like Salesforce, Adobe, Disney, and Amazon. Zao has a rich history in the tech industry, co-creating Farmville (200 million users, $3B revenue) and Kaspa ($3B market cap). Self-taught in deep learning, generative AI, and machine learning, Zao is passionate about empowering others to develop practical AI applications. His extensive knowledge of both the technical and business sides of AI projects will be invaluable as you work on your own.

With Alvin and Zao's guidance, you’ll gain practical insights into building and deploying advanced AI models, preparing you for the most challenging and rewarding roles in the AI field.

You’ll receive a comprehensive set of resources to help you master large language models.

Resources

Unlock exclusive bonuses to accelerate your AI journey.

Bonus

Everyone’s heard of ChatGPT, but what truly powers these modern large language models? It all starts with the transformer architecture. This bootcamp demystifies LLMs, taking you from concept to code and giving you a full, hands-on understanding of how transformers work. You’ll gain intuitive insights into the core components—autoregressive decoding, multi-head attention, and more—while bridging theory, math, and code. By the end, you’ll be ready to understand, build, and optimize LLMs, with the skills to read research papers, evaluate models, and confidently tackle ML interviews.

AI Bootcamp

Be able to build large language models, which can increase your salaries by $50k a year. Worth $500k over 10 years

Cheatsheet on generative AI interviews for FANGs, a $50k a year over a $500k value

A complete course on end to end streaming Langchain with a fully functional application for startups. $15k in value

Be able run consulting in AI $100k in annual value. Over 10 years. $1m

Be able to build an AI company $1M in annual value

Technical and business design review from Alvin and Zao about your project. $25000 dollars in value

$3.4M in value. Only for $8k. Available now for $7000 dollars in pre-payment. Payment plan for $2667 per month. This will be a $10k to $15k bootcamp in the future

What You will Gain

Understand the lifecycle of large language models, from training to inference

Build and deploy a fully functional LLM Inference API

Master tokenization techniques, including byte-pair encoding and word embeddings

Develop foundational models like n-grams and transition to transformer-based models

Implement self-attention and feed-forward neural networks in transformers

Evaluate LLM performance using metrics like perplexity

Deploy models using modern tools like Huggingface, Modal, and TorchScript

Adapt pre-trained LLMs through fine-tuning and retrieval-augmented generation (RAG)

Leverage state-of-the-art tools for data curation and adding ethical guardrails

Apply instruction-tuning techniques with low-rank adapters

Explore multi-modal LLMs integrating text, voice, images, and robotics

Understand machine learning operations, from project scoping to deployment

Design intelligent agents with planning, reflection, and collaboration capabilities

Keep up-to-date with AI trends, tools, and industry best practices

Receive technical reviews and mentorship to refine your projects

Create a robust portfolio showcasing real-world AI applications

Learn

The newline Guide to Building Your First GraphQL Server with Node and TypeScript

Teach

Amelia Wattenberger

Author of Fullstack D3

Community

Masterclasses

Tutorials

Fullstack React with TypeScript