4 hands-on projects to grow as an AI Engineer
From building WhatsApp Agents and Finetuning LLMs to messing with Vector Search
Like a philosopher once said (okay, it was
): “You are what you eat”.So today, I’m serving up 4 spicy little dishes. Each one’s got its own flavor; some sweet, some savory, maybe even a little crunchy or “schwifty” (Rick & Morty reference). You’ll probably have a favorite (I do), but hey, try them all.
Chef’s recommendation 👨🍳
The best part? These arent’s empty-calorie, all-sugar “AI Influencer” snacks. Nah. These are protein-packed meals that’ll help you grow as a real ML / AI Engineer. Because around here, we build stuff - we don’t just repost GPT screenshots and call it thought leadership
Alright, enough with the food talk.
Let’s eat! 🍽️
Dish 1: Agent Design Patterns
🌶️ Spice Level: Medium-Hot
💻 Check the code here!
This is the first “open-source” course I’ve ever released, and it holds a special place in my heart because of that.
It’s structured in 4 lessons, where you’ll learn how to implement, from scratch, the 4 agentic patterns as defined by Andrew Ng (check this video):
And when I say “from scratch,” I really mean it - no LangChain, no LlamaIndex, no CrewAI. Just good old Python and Groq LLMs. Keeping it simple.
Each lesson comes with its own notebook, Python module, video, and a Substack article to tie it all together.
If you’re feeling brave, there’s also a full version that covers all the patterns in one go (fair warning: it’s 1 hour and 41 minutes long 😅).
Dish 2: Rick LLM
🌶️🌶️ Spice Level: Hot
💻 Check the code here!
The idea is pretty simple: create a local finetuned LLM that talks like Rick Sanchez.
It’s the perfect course (or dish?) for anyone looking for a more... unconventional way to learn about LLM finetuning. The project’s broken down into three main phases:
Dataset creation - Where I’ll teach you how to create custom Instruct Datasets from Rick and Morty transcripts.
Model finetuning - Finetuning Llama 3.1 8B using Unsloth on Lambda Labs GPUs.
Model deployment - Converting and deploying the model to Ollama for local use.
Besides the repo, this one’s got its own video too - with what’s probably my best thumbnail yet 😂
There’s also a written version of the whole project over on Substack, if you prefer reading.
Dish 3: Twin Celebrity App
🌶️ Spice Level: Mild
💻 Check the code here!
Ever wondered which celebrity you look like the most?
This project answers that burning question by building the architecture shown in the diagram above. You’ll be putting together two main components:
Embedding Generation Pipeline - Handles extracting face embeddings from celeb images using the FaceNet model and storing them in Qdrant. It’s all orchestrated with ZenML, so you’ll also get some hands-on time with that framework.
Streamlit Application - This is the front end where users can upload their photo, find their celeb twin, and see the results.
And yeah, I know what you’re thinking: “Who’s Miguel’s celeb lookalike? Brad Pitt? George Clooney? Keanu Reeves?”
Nope. It’s Andy Samberg. Reality check for Miguel 😅
If this project sounds like your kind of flavor, go check out the video too.
There’s also a Substack article for this one, if you’re more into reading (or just wanna scroll through it while sipping coffee and looking productive).
Dish 4: Ava, the WhatsApp Agent
🌶️🌶️🌶️ Spice Level: Nuclear
💻 Check the code here!
I’ll keep this one short, since Jesús Copado and I have been chatting about it nonstop for weeks now 😅
In a nutshell, you’ll learn how to build a multimodal agent connected to WhatsApp that talks and reacts like a real person. Images, voice messages … you name it. Ava can handle it all.
Oh, and there’s a full course available too (yep, over 2 hours!) that dives into every part of the project. Go check it out!
Dish 5: ?
What? But Chef, didn’t you say 4 dishes?
I did… but there’s one more cooking as we speak. Almost ready to be served. And trust me - it’s gonna be very different from the others.
This one’s made with the help of a very special co-chef … but sorry, can’t reveal the name just yet.
You’ll have to wait until this Thursday for the big reveal.
Until then … bon appétit, friends! 🍽️
Dish2 and Dish3 code link unavailable
What software can be used to draw such beautiful diagrams?