course-AIEngineers2025 by Nutchanon Yongsatianchot

interesting.

cluster: ai_and_cs_related_courses

general knowledge about this class

• focus more on practical than research
  • course-generative_ai2024_by nutchanon_yongsatianchot is more research oriented
• from the first lecture, this course focuses on
  • yes
    • Dev with AI (vibe coding, AI Agents, API)
    • LLM stuffs (RAG, Knowledge Agent)
  • nope
    • DL
    • ML Engineer
    • Data Science
  • aim for final project with NSC level
  • student can build their AI augmented software (game, etc.)

Lec01

My takeaways from this Lec

• this course focuses on
  • yes
    • Dev with AI (vibe coding, AI Agents, API)
    • LLM stuffs (RAG, Knowledge Agent)
  • nope
    • DL
    • ML Engineer
    • Data Science
  • aim for final project with NSC level
  • student can build their AI augmented software (game, etc.)
• must look up
  • book-AI_for_highschool
  • alexnet
  • 2017 Attention is All You Need
  • from palm, language translation
    • https://research.google/blog/pathways-language-model-palm-scaling-to-540-billion-parameters-for-breakthrough-performance/
    • https://blog.mojeek.com/2023/11/one-model-to-rule-them-all.html
• Andrej Karpathy: interesting guy
• Autonomy slider
• token, tokenizer, Auto Regressive Decoding, context_window
• LMArena, openrouter

AI history

• 195x AI
• perceptron
• ELIZA
• Deep Learning 2012 AlexNet(1M perceptron), ImageNet challenge
  • DL with GPU
• 2017 Attention is All You Need
• from palm, language translation
  • https://research.google/blog/pathways-language-model-palm-scaling-to-540-billion-parameters-for-breakthrough-performance/
  • https://blog.mojeek.com/2023/11/one-model-to-rule-them-all.html
• 2022 Nov: ChatGPT

music gen tools

• suno
• udion
• deepmind lyria
• elevenlabs.io musics

world model

google genie 3

AI Builder Tech Stack

https://www.iconiqcapital.com/growth/reports/2025-state-of-ai

• model training & finetuning
• llm & ai application development
• monitoring & observability
• inference optimization
• model hosting
• model evaluation
• data processing & feature engineering
• vector database
• synthetic data & data augmentation
• coding assistant
• devops & mlops
• product & design

Andrej Karpathy: Software Is Changing (Again)

1.19 actions for LLMs (prompt friendly)

• web with llm.txt or robot.txt
• tool: gitingest

AI Engineer

1:20:00

• LLM Engineer
• workflow compared to ML Engineer
  • ML: Data -> Model -> Product
  • AI: Product ->(if success) Data ->(if scaling) Model
• AI Engineer World’s Fair
• GenAI Application Engineers
  • Prompting techniques
  • Agentic Frameworks
  • Guardrails
  • Evals
  • RAG
  • Fine-tuning
  • Computer use
  • Asynchronous programming
• twitter: Sri Laasya Nutheti: Building AI agents is 5% AI and 100% software engineering

Basic NLP

• 1:31:24
• token
  • represent word, word part, special character
  • number
• tokenizer
  • model performance, delulu: this one’s duty
• P(next token|“I want to”)
• Auto Regressive Decoding (traditional way to predict next word until stop token)
  • compute probability over the next token
  • select next token
    • temp
    • sample over top n
  • append the selected token to the context
  • repeat until stop token
• context_window
  • similar to working memory
  • input size
  • total no. of token that the model can consider at once

Find interesting models

• big three: openai anthropic gemini
• LMarena
• openrouter: use many models

others

• 59.37: MLX
• learn AI Engineering with textbook: outdated very fast
• recon this
  • https://www.anthropic.com/learn

Lec02

Lec03

Lec04

Lec05

Lec06

Lec07

Lec08

Lec09

Lec10

Lec11

Lec12

created on: Thu Oct 02 2025