Akshay Pachaar

Stanford offers FREE Data Science Education!

World-class courses on:

- Machine Learning
- Computer Vision
- Statistical Learning
- Graph Neural Networks
- Reinforcement Learning
- Natural language processing (NLP)

Read more 🧵👇

1️⃣ Machine Learning

The best Machine Learning course out there.

Taught by @AndrewYNg, Tested by time & taken by millions.

A great place to start! 🚀

Check this out👇
https://lnkd.in/dY9ZXJyM

2️⃣ Databases: Relational Databases & SQL

This course provides an introduction to relational databases and comprehensive coverage of SQL.

Standard query language for relational database systems.

Check this out👇
https://lnkd.in/dNQMknrQ

3️⃣ Statistical learning

Learn some of the main tools used in statistical modeling and data science.

Covers both traditional as well as exciting new methods.

Check this out👇
https://lnkd.in/dUE5jkXr

4️⃣ Computer vision

A great course covering fundamentals of Deep Learning, focusing on Computer vision applications!

Its was started by Andrej Karpathy at Stanford!

Check this out👇
https://lnkd.in/dhWrkM5P

4️⃣ NLP

A thorough introduction to cutting-edge neural networks for NLP.

A hands-on course with plenty of assignments & a final project.

PyTorch for all the code! 💙

Check this out👇
https://lnkd.in/dUKWjMJH

6️⃣ Reinforcement learning

This class will provide a solid introduction to the field of reinforcement learning.

You'll learn through a combination of lectures, & written and coding assignments.

Check this out👇
https://lnkd.in/dQRMqean

7️⃣ ML on Graphs

This course covers important research on the structure & analysis of large social & information networks.

I have personally learnt a lot from this course & it helped me solve problems where traditional methods failed.

Check this out👇
https://lnkd.in/d4C44VRy

That's a wrap!

If you interested in:

- Python 🐍
- Data Science 📈
- Machine Learning 🤖
- Maths for ML 🧮
- MLOps 🛠
- CV/NLP 🗣
- LLMs 🧠

Find me → https://lnkd.in/em_V4unu ✔️
Everyday, I share tutorials on above topics!

Check my tutorial on Self-Attention👇
https://lnkd.in/dwVjZXKU

Top 5 Agentic AI design patterns, visually explained!
.
.
Agents allow a LLMs to repeatedly refine their output by incorporating self-evaluation, planning, and collaboration!

Today, we'll understand some popular design patterns employed in building with AI agents.

Let's go! 🚀

1️⃣ Reflection Pattern

In this pattern, the AI “reviews” its own work—spotting mistakes and iterating until the final outcome is polished.

Think of it like a self-critique cycle where the model refines each draft before presenting the result!

2️⃣ Tool Use Pattern

Imagine the tool use agent as a person, with the LLM as its brain and a set of tools as its hands to take actions.

By querying databases, calling APIs, and executing Python scripts, the LLM moves beyond relying solely on its internal knowledge.

3️⃣ ReAct (Reasoning and Acting)

ReAct combines Reflection and Tool use patterns.

Now the agent can reflect on both the LLM-generated outputs and the results of its interactions with the external world using tools.

One of the most popular and powerful patterns used today.

4️⃣ Planning Pattern

This is all about task breakdown & strategic thinking.

Instead of tackling a big request in one go, the AI creates a roadmap:

- subdivides tasks
- outlines objectives

Just like a project manager organizing a complex workflow.

5️⃣ Multi-Agnet Pattern

Multiple specialized agents working together—each with its own role!

Imagine an engineering team, the PM sets goals, the Tech Lead designs, the Developer codes, and the DevOps Engineer handles deployment.

Together, they deliver a unified product!

I hope you enjoyed this!

👉 Want to master Agentic AI, I've have put together a hands-on course on Building with AI agents, no pre-requisites everything you need in one place. Link in the comments.
_____
Share this with your network if you found this insightful ♻️
Follow me (Akshay Pachaar) for more insights and tutorials on AI and Machine Learning!

𝗧𝗵𝗶𝘀 𝗣𝘆𝘁𝗵𝗼𝗻 𝗱𝗲𝗰𝗼𝗿𝗮𝘁𝗼𝗿 𝗶𝘀 𝘀𝗼 𝗽𝗼𝘄𝗲𝗿𝗳𝘂𝗹! 🔥

Introducing 𝗟𝗮𝘁𝗲𝗫𝗶𝗳𝘆! 🚀

Now you can easily generate mathematical descriptions for python functions & document your code! 🚀

🔵 Jupyter Notebook 📒 ⬇️
https://lnkd.in/dPa8HC-7

if you're interested in:

- Python 🐍
- Maths for ML 🧮
- ML/MLOps 🛠
- CV/NLP 🗣
- LLMs 🧠
- AI Engineering ⚙️

Find me → https://lnkd.in/em_V4unu ✔️
Everyday, I share tutorials on above topics!

Here's LateXify for you! 🚀👇

4 ways to run LLMs like DeepSeek-R1 locally on your computer!
.
.
Running LLMs locally is like having a superpower:

- Cost savings
- Privacy: Your data stays on your computer
- Plus, it's incredibly fun

Today, we'll explore some of the best methods to achieve this.

Let's go! 🚀

1️⃣ Ollama

Running a model through Ollama is as simple as executing a command:

ollama run deepseek-r1

You can also install Ollama with a single command:

curl -fsSL https:// ollama. com/install .sh | sh

Check this out👇
ollama.com

2️⃣ LMStudio

LMStudio can be installed as an app on your computer.

It offers a ChatGPT-like interface, allowing you to load and eject models as if you were handling tapes in a tape recorder.

Check this out👇
lmstudio.ai

3️⃣ vLLM

vLLM is a fast and easy-to-use library for LLM inference and serving.

It provides State-of-the-art serving throughput ⚡️

A few lines of code and you can locally run DeepSeek as an OpenAI compatible server with reasoning enabled.

Check this out👇
https://lnkd.in/gCZK9qUz

4️⃣ LlamaCPP (the OG)

Georgi Gerganov's LlamaCPP enables LLM inference with minimal setup and state-of-the-art performance.

Check this out👇
https://lnkd.in/gtVniS3q

Interested in ML/AI Engineering? Sign up for our newsletter for in-depth lessons and get a FREE eBook with 150+ core DS/ML lessons: https://lnkd.in/gB7yHExC
_____
Find me → Akshay Pachaar ✔️
For more insights and tutorials on AI and Machine Learning.

Harvard university is offering FREE world class education in Data Science!

Courses cover:
- Python
- Data Visualization
- Probability
- Statistics
- Machine Learning
- Data Science: Capstone

A project-based pedagogy that allows you to learn while building! 🚀

Read more 🧵👇

1️⃣ CS50p: Python

If you are new to programming and just getting started.

There isn't a better place to learn Python than @davidjmalan 's CS50p.

Beautiful explanations and great projects.
It's a complete package.

Check this out 👇
https://lnkd.in/enGx-BDY

2️⃣ Data Visualization

Learn basic data visualization principles and how to apply them using ggplot2.

Check this out👇
https://lnkd.in/ecHiPHVz

3️⃣ Probability

Learn probability theory essential for a data scientist, using a case study on the financial crisis of 2007-2008.

A highly practical approach!

Check this out👇
https://lnkd.in/e8YrS24y

4️⃣ Statistics

Learn inference and modeling, two of the most widely used statistical tools in data analysis.

Check this out👇
https://lnkd.in/ekmQGsqX

5️⃣ Machine Learning

Starts from the basics!

And, you'll get to build a movie recommendation system, one of the most popular and successful data science techniques.

Check this out👇
https://lnkd.in/eSGsqM3H

6️⃣ Data Science: Capstone

When you complete this project you will have a data product to show off to potential employers.

A strong indicator of your expertise in the field of data science.

Check this out👇
https://lnkd.in/eizVpmU6

That's a wrap!

If you interested in:

- Python 🐍
- AI/ML 🤖
- Maths for ML 🧮
- MLOps 🛠
- CV/NLP 🗣
- LLMs 🧠

Find me → https://lnkd.in/em_V4unu ✔️
Everyday, I share tutorials on above topics!

Check one tutorial on DBSCAN👇
https://lnkd.in/dnB2zZay

Mixture of Experts vs. Transformers, explained visually:

(a popular LLM interview question)

Mixture of Experts (MoE) is a popular architecture that uses different “experts“ to improve Transformer models.

As shown in the visual below, Transformer and MoE mainly differ in the decoder block:

- Transformer uses a feed-forward network.
- MoE uses experts, which are feed-forward networks but smaller compared to that in Transformer.

During inference, a subset of experts are selected. This makes inference faster in MoE.

Since the network has multiple decoder layers:
- the text passes through different experts across layers.
- the chosen experts also differ between tokens.

But how does the model decide which experts should be ideal?

The router does that.

The router is like a multi-class classifier that produces softmax scores over experts. Based on the scores, we select the top K experts.

The router is trained with the network and it learns to select the best experts.

But it isn't straightforward.

There are challenges!

Challenge 1) Notice this pattern at the start of training:

- Say the model selects “Expert 2“
- This expert gets a bit better
- It may get selected again
- The expert learns more
- It gets selected again
- It learns more
- And so on!

Essentially, a few experts could be over-exposed to training while many experts may go under-trained!

We solve this in two steps:

- Add noise to the feed-forward output of the router so that other experts can get higher logits.
- Set all but top K logits to -infinity. After softmax, these scores become zero.

 This way, other experts also get the opportunity to train.

Challenge 2) Some experts may get exposed to more tokens than others—leading to under-trained experts.

We prevent this by limiting the number of tokens an expert can process.

If an expert reaches the limit, the input token is passed to the next best expert instead.

In terms of parameters, MoEs have more parameters to load. However, a fraction of them are activated since we only select some experts.

This leads to faster inference. Mixtral 8x7B is one famous LLM that is based on MoE.

Over to you: Do you like the strategy of multiple experts instead of a single feed-forward network?
____
👉 Interested in ML/AI Engineering? Sign up for our newsletter for in-depth lessons and get a FREE eBook with 150+ core DS/ML lessons: https://lnkd.in/gB7yHExC
_____
Share this with your network if you found this insightful ♻️
Follow me (Akshay Pachaar) for more insights and tutorials on AI and Machine Learning!

This Python decorator is so powerful! 🔥

Introducing LateXify! 🚀

Now you can easily generate mathematical descriptions for python functions & document your code! 🚀

🔵 Jupyter Notebook 📒 ⬇️
https://lnkd.in/dPa8HC-7
___________________
if you're interested in:

- Python 🐍
- Machine Learning 🤖
- AI Engineering ⚙️

Find me → https://lnkd.in/em_V4unu ✔️
I also write a Newsletter on AI Engineering, Join 8k+ reader: https://lnkd.in/gRQX28B8
___________________
Here's LateXify for you! 🚀👇