I’ve been working in AI for close to a decade, starting well before today’s LLM-driven wave.

My work is grounded in understanding AI across layers — from hardware, AI models, and data to system design, reliability, and scale. I approach AI as an end-to-end engineering problem: not just modeling, but how models interact with data, infrastructure, and real-world constraints.

Alongside this, I write The AI Merge, a technical newsletter on AI foundations and systems engineering, focused on clear thinking and practical system design rather than hype.

I’m currently building an AI Systems course, centered on end-to-end system design, for engineers who want a deeper, systems-first understanding of modern AI.

🗞️ Read The AI Merge: https://read.theaimerge.com

Alex Razvant

Read more code.

And fewer articles, tutorials, or surface-level books.

If you’re already a Software Engineer or AI / ML Engineer, one of the fastest ways to level up right now is to: 

Study how big AI teams write and ship code - in public. 

Many top companies and major contributors to the AI space have been open-sourcing their internal frameworks, kernels, or infrastructure libraries. 

You might feel overwhelmed by the rapid pace at which AI is going.
But this is a massive learning opportunity if you know where to look.

Learning doesn’t just happen from papers and tutorials, especially if you already build software.

It happens when you:
- Read real production code
- Study PRs or commit history
- See how teams structure repos, APIs, configs, tests

Check what gets reviewed, what gets rejected, and why.
This kind of learning will upgrade your engineering intuition. 

Here are some excellent codebases to learn from.

✅ From DeepSeek:
1. FlashMLA  - A library of high-performance attention kernels.
Link: https://buff.ly/XbFnPA7 

2. DeepEP - A Communication Library for high-throughput MoE workloads.
Link: https://buff.ly/3qxMc3S

3. 3FS - A Distributed Filesystem for AI Workloads, for AI training and inference.
Link: https://buff.ly/ovfW47A

✅ From NVIDIA:
1. Dynamo - A datacenter-scale Inference Serving Framework for GenAI workloads.
Link: https://buff.ly/WyAMzBz 

2. NeMo  -  A Scalable Generative AI Framework, for training and deploying LLMs and VLMs.
Link: https://buff.ly/L9KTuJp 

✅ From vLLM:
1. The vLLM Library  - One of the most popular LLM inference engines, designed for scale.
Link: https://buff.ly/GLCfoIM

✅ From SGLang:
1. The SGLang Engine - A high-performance serving engine for LLMs & multimodal models.
Link: https://buff.ly/m4HhsWQ 

✅ From llama.cpp
1. The llama.cpp Engine - A minimal C/C++ inference engine for LLMs with broad hardware support and quantization options.
Link: https://buff.ly/u9yaXwp

📌 Don't study everything, you won't be able to.

Instead, pay attention to:
⇾ Code structure & abstractions
How are core functionalities separated (core logic vs hardware-specific optimizations)?

⇾  PRs & commit history
What context do authors provide? How do reviewers push back?

⇾  Interfaces & public APIs
How are public APIs designed?

⇾  Tests & benchmarks
How do the test suites look? What CI pipelines do they have, profiling? 

Don’t just use these libraries. 
Read them.

That will build your intuition and help you get better at your craft, either as a curious learner or an engineer as part of a larger team at your job.

---
I distill my learnings with my audience on The AI Merge.
Check it here (subscribe if interested) https://buff.ly/2rwqlNL

Senior AI Engineer | Writing The AI Merge Newsletter | Helping Engineers Build AI Beyond Demos

CarbabaAI, a UAE-based B2B SaaS company, pioneers document intelligence using AI-driven compliance systems and QA validators to optimize enterprise workflows. As Founder, contributions include innovative solutions that reduced document processing time by 40%, showcasing a commitment to operational efficiency and transformative technologies.  

With a biomedical degree major with a data science minor from Swinburne University of Technology, expertise spans machine learning, AI systems, and data analytics. Previous roles, such as Software Developer at CliniNote, focused on real-world data solutions for healthcare. Additionally, mentoring at Monash University bridged academic concepts with practical applications, reflecting a dedication to responsible AI-driven industry innovation.

Ghadeer A.

Why 90% of ML Engineers Quit Before Year 2 (SAVE THIS 👇)

It doesn’t have to be this way

Most learners drown in models with no map.
Nothing sticks.

Concepts blur together.
Progress stalls fast.

A full visual system fixes everything.
Clarity arrives instantly.  

✅ This collection maps Neural Networks end-to-end.
Page 1 shows every major architecture in one view.
You see how each model connects across the ecosystem.

✅ The Neural Network Graphs on page 2 break down flow.
They show feed-forward, recurrent, LSTM and GRU paths.
Patterns become visible instead of abstract.

✅ Pages 3–4 decode the maths behind learning.
Not theory for theory’s sake. Practical intuition.
You finally understand why models behave as they do.

✅ Machine Learning in Emoji on page 5 simplifies concepts.
Complex ideas become instantly memorable.
Supervised, unsupervised and reinforcement made visual.

✅ Scikit-Learn, Azure ML and Python basics span pages 6–10.
They show algorithms, workflows and starter code together.
Zero guesswork on what to use and when.

✅ Pandas, NumPy and data wrangling fill pages 11–15.
You see cleaning, shaping, merging and plotting clearly.
Data stops being messy and starts being manageable.

✅ SciPy, Matplotlib and ggplot2 visuals on pages 16–18.
They reveal analysis, plotting and scale choices.
Your data storytelling sharpens instantly.

✅ PySpark and Big-O on pages 19–20 zoom out further.
You understand scalability and operations at a glance.
It becomes the roadmap many engineers never get.

This is the satellite view.
This is the missing map.
This is how AI finally clicks.
This is how you build intuition fast.

➕ Follow Ghadeer for more insights
♻️ Repost to help others in your network
📩 Save for later

Ghadeer  A.

Document-to-Decision Architect | Building Enterprise AI Systems that Turn Documents into Trusted Decisions

Neural_Networks_Guide

🚨 BREAKING: I wasted $2,000 learning AI the wrong way.
Turns out NVIDIA has a free beginner path. WITH certificates.

You don’t need a CS degree.
You don’t need a GPU.
You don’t need paid courses.

You just need the right order 👇

STEP 1 — Understand GenAI (no coding)
Generative AI Explained
What LLMs are, how GenAI works, where it’s used.
🔗 https://lnkd.in/dCy_GBct

STEP 2 — Build your first real AI system
Augment Your LLM Using RAG
Learn how AI reads PDFs, docs & private data.
🔗 https://lnkd.in/d6JebWSx

STEP 3 — Learn how AI “thinks”
Building a Brain in 10 Minutes
Neural networks explained visually. No math overload.
🔗 https://lnkd.in/dhuRZ3XP

STEP 4 — (Optional) Understand why GPUs matter
An Even Easier Introduction to CUDA
Why NVIDIA GPUs unlocked modern AI.
🔗 https://lnkd.in/dX4k6fdi

STEP 5 — Make it career-ready
• Finish labs
• Add certificate to LinkedIn
• Build 1 tiny demo (PDF → AI answers)

💡 Beginner rule:
Don’t binge courses.
Finish one → build one → share one.

➕ Follow Ghadeer A.  for more insights
♻️ Repost to help others in your network
📩 Save for later

No alternative text description for this image

The biggest lie in AI engineering right now?
“We built an agent.”

We’re asking agents to reason, plan, adapt, and act autonomously but most teams are stitching together prompts and hoping it holds.

That’s why I wanted to share the stack I keep coming back to.

The one that actually survives messy workflows, inconsistent inputs, and real users.

Here’s the Agentic RAG stack behind the AI systems that don’t just run they improve 👇

0️⃣ Deployment
Where your agent lives.
Stable. Scalable. Ready for real load.

1️⃣ Evaluation
Tight feedback loops.
No evaluation → no improvement.

2️⃣ LLMs
The thinking layer.
Reasoning. Planning. Decision-making in natural language.

3️⃣ Frameworks
The wiring of the system.
Prompts, memory, tools orchestrated instead of duct-taped.

4️⃣ Vector Databases
The context engine.
Fast retrieval → relevant responses.

5️⃣ Embeddings
Turn text into meaning.
Make knowledge usable.

6️⃣ Data Extraction
Web. Docs. APIs.
Your agent is only as informed as its inputs.

7️⃣ Memory
The part that makes your agent feel alive.
Short-term + long-term + episodic context.

8️⃣ Alignment
Safety, telemetry, and trust.
You can’t scale what you can’t control.

I’ve seen this stack turn fragile prototypes into reliable systems.
And once teams get it right, something shifts:

They stop building demos.
They start building intelligence.

What part of the agent stack are you wrestling with right now?

➕ Follow Ghadeer for more insights
♻ Repost to help others
📩 Save for later

🚨BREAKING: Anthropic just released ALL the Claude Code secrets

Claude Code feels powerful, but unpredictable.

Hallucinations, wasted tokens, messy edits.

Anthropic just revealed how to fix all of it.

Here are the 7 secrets that make Claude Code elite 👇:

1️⃣ Use GitHub immediately

Connect Claude Code to Git early.
It reads your full commit history.
Hallucinations drop dramatically.

2️⃣ Load as much context as possible

Context handling was massively upgraded.
Claude now compacts context intelligently.
You can keep projects alive for a long time.

3️⃣ Control eagerness

Claude Opus 4.5 is extremely eager.
It acts first and apologizes later.
Guide it clearly to avoid chaos.

4️⃣ Avoid the word “think”

“Think” triggers extra token usage.
So do “think more” and “ultrathink.”
Use “evaluate” or “consider” instead.

5️⃣ Exploit vision capabilities

Opus 4.5 dominates visual reasoning.
It understands images better than any model.
Paste images directly for debugging or ideas.

6️⃣ Run tools in parallel

Claude is the best tool-using model.
Anthropic added a new parallel rule.
Tools now run faster and smarter together.

7️⃣ Actively reduce hallucinations

Eagerness causes premature file edits.
Claude may act without full understanding.
Force research before execution.

These practices change everything.
Claude becomes calmer, cheaper, smarter.
And dramatically more reliable.

Link to full prompt guide: https://lnkd.in/d8rqisP2

➕ Follow Ghadeer for more insights
♻️ Repost to help others in your network
📩 Save for later

AI doesn’t understand anything. Here’s exactly what it does instead 👇

Most people think LLMs “understand” language.
They don’t.

They predict the next token INSANELY well.

This video makes it click.

You’re watching text get:
→ chopped into pieces
→ turned into math
→ weighed against everything else
→ refined layer by layer
→ sampled into words that feel intelligent

No thoughts.
No memory of meaning.
Just probability + attention at scale.

Once you see this, two things change:
1️⃣ You stop mystifying AI
2️⃣ You start engineering it

If you’re building with LLMs and this isn’t crystal clear yet you’re flying blind.

Watch this slowly.
Pause it.
Rewatch it.

➕ Follow Ghadeer for more insights
♻️ Repost to help others in your network
📩 Save for later

🚨I spent the last 2 weeks reading 50+ Claude Code articles, X threads, and “guides.” (SAVE THIS 👇)

Most of them were AI slop.
Recycled tips. Vague workflows. Zero leverage.

But the top 1% completely changed how I build with Claude Code.

So I compiled the only Claude Code guide you actually need to 10× your workflows.

Your Claude Code workflows are about to get 10× better

This is a curated crash course for builders who want:
• Faster iteration
• Cleaner vibe-coded workflows
• Less steering, more output

If you use Claude Code seriously, bookmark this.

How to use this guide (important)

If you’re new → spend more time in Getting Started
If you’re advanced → jump to Pro Tips & Resources

Pro move:
Create a new Claude project and paste these resources into it.
Treat Claude as a tutor + co-builder, not a chatbot.

1️⃣ Getting Started

Start here if you’re new:

• Claude Code in <15 Minutes
Setup Claude Desktop, Web, CLI, editor tools, terminal, extensions
👉 https://lnkd.in/dY7wGJFh

• Elite Claude Code Setup (Boris Cherny) 
This alone fixes 80% of bad CC usage

• Free Claude Code Basics Course (3+ hours)
Structured, no fluff
👉 https://lnkd.in/dUCxC4YP

2️⃣ Pro Tips / Best Practices (must-read)

• Official Claude Code agentic coding best practices
👉 https://lnkd.in/d__wJwMc

• Hidden power tips (criminally underrated)
👉 https://lnkd.in/dWG_xFvR

These two alone will level up your planning, context, and tool usage.

3️⃣ Builder Tools (for real leverage)

Once you have the basics down, this is where speed comes from.

I’ve collected:
• The single best Claude Code prompt
• 12 best Claude Code prompts I use daily

This should honestly be paid.

But I’m giving it away.

👇
Comment “PROMPTS” and I’ll send you the full list.

Thank me later.

➕ Follow Ghadeer for more insights
♻️ Repost to help others in your network
📩 Save for later

The sexiest job of 2026 isn’t AI Engineer. It’s Forward Deployed Engineer.

Here’s the real reason why:

Saw this from Eduardo Ordax and it made me pause.

Not because it’s clever.
Because it’s accurate.

We keep talking about smarter models.
But the real bottleneck isn’t intelligence anymore.

It’s translation.

Most companies don’t fail at AI because they lack talent.
They fail because no one owns the space between what leadership wants and what systems can actually do.

That gap is where Forward Deployed Engineers live.

They’re not there to invent.
They’re there to confront reality.

To turn:
– ambition into constraints
– demos into workflows
– “AI strategy” into something that survives production

Once you notice it, you see this role everywhere.

Not as a title.
As a pattern.

In 2026, the most valuable people in AI won’t be the loudest or the most technical.

They’ll be the ones who can walk into confusion…
…and quietly make things work.


Bookmark this 🔥. You’ll see this job title everywhere this year.

➕ Follow Ghadeer for more insights
♻️ Repost to help others in your network
📩 Save for later

🚨 BREAKING: 200k tokens doesn’t mean you have context.

It means you have a budget.

Claude didn’t suddenly get worse.
Your outputs didn’t magically decay.

Your context did.

Context = everything the model sees when it responds:
• your prompt
• prior messages
• system instructions
• tools & agents
• metadata

Most people treat this like free real estate.

It’s not.

Hard truth:
More context ≠ better output.
Past a point, model performance drops.

That’s where “AI slop” comes from.

Not bad models.
Bloated context.

What actually works 👇

• One clear objective per conversation
• Plan before execution (force the model to ask questions)
• Eliminate ambiguity early
• Reset bad threads instead of “fixing” them
• Pull docs/tools only when needed
• Offload heavy research to sub-agents
• Optimize for signal, not clever setups

If a token doesn’t help answer the next question,
it’s already hurting you.

Context hygiene > prompt tricks.

👇 Comment “CONTEXT” and I’ll share the full 7-step context engineering playbook.

➕ Follow Ghadeer for more insights
♻️ Repost to help others in your network
📩 Save for later

taplio

The Battle

Alex Razvant

Ghadeer A.

Followers

Posts

Engagements

Likes

Performance ScoreAlex Razvant's LinkedIn Performance

Performance ScoreGhadeer A.'s LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Alex Razvant

Ghadeer A.

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Alex Razvant's LinkedIn Performance

Performance ScoreGhadeer A.'s LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreAlex Razvant's LinkedIn Performance