Amir Zeinali
@amir-zeinali.bsky.social
Machine Learning Consultant | AI/ML, GenAI, MLOps | Problem Solver
CEOs: “Why should I hire a consultant when my team knows ML?”
Here’s my simple analogy:
Your team might be great at building a car, but does it know how to engineer a racetrack?
Consultants bring a system-level view:
we connect models, workflows, and business strategy.
Here’s my simple analogy:
Your team might be great at building a car, but does it know how to engineer a racetrack?
Consultants bring a system-level view:
we connect models, workflows, and business strategy.
January 22, 2025 at 9:59 PM
CEOs: “Why should I hire a consultant when my team knows ML?”
Here’s my simple analogy:
Your team might be great at building a car, but does it know how to engineer a racetrack?
Consultants bring a system-level view:
we connect models, workflows, and business strategy.
Here’s my simple analogy:
Your team might be great at building a car, but does it know how to engineer a racetrack?
Consultants bring a system-level view:
we connect models, workflows, and business strategy.
DeepSeek-AI has released its latest breakthrough in reasoning for Large Language Models (LLMs):
DeepSeek-R1.
This innovative model introduces a multi-stage pipeline leveraging Reinforcement Learning (RL) to tackle complex reasoning tasks without extensive supervised fine-tuning.
DeepSeek-R1.
This innovative model introduces a multi-stage pipeline leveraging Reinforcement Learning (RL) to tackle complex reasoning tasks without extensive supervised fine-tuning.
January 21, 2025 at 11:37 AM
DeepSeek-AI has released its latest breakthrough in reasoning for Large Language Models (LLMs):
DeepSeek-R1.
This innovative model introduces a multi-stage pipeline leveraging Reinforcement Learning (RL) to tackle complex reasoning tasks without extensive supervised fine-tuning.
DeepSeek-R1.
This innovative model introduces a multi-stage pipeline leveraging Reinforcement Learning (RL) to tackle complex reasoning tasks without extensive supervised fine-tuning.
✴️ Awesome Artificial Intelligence: an awesome repo for AI engineers
It’s packed with AI tools, courses, books, lectures, and papers.
Perfect for both aspiring and seasoned AI engineers.
Big thanks to Owain Lewis for curating this must-see collection.
🔗 github.com/owainlewis/a...
It’s packed with AI tools, courses, books, lectures, and papers.
Perfect for both aspiring and seasoned AI engineers.
Big thanks to Owain Lewis for curating this must-see collection.
🔗 github.com/owainlewis/a...
GitHub - owainlewis/awesome-artificial-intelligence: A curated list of Artificial Intelligence (AI) courses, books, video lectures and papers.
A curated list of Artificial Intelligence (AI) courses, books, video lectures and papers. - owainlewis/awesome-artificial-intelligence
github.com
January 19, 2025 at 9:14 PM
✴️ Awesome Artificial Intelligence: an awesome repo for AI engineers
It’s packed with AI tools, courses, books, lectures, and papers.
Perfect for both aspiring and seasoned AI engineers.
Big thanks to Owain Lewis for curating this must-see collection.
🔗 github.com/owainlewis/a...
It’s packed with AI tools, courses, books, lectures, and papers.
Perfect for both aspiring and seasoned AI engineers.
Big thanks to Owain Lewis for curating this must-see collection.
🔗 github.com/owainlewis/a...
During a project for a VP of Engineering, we faced a familiar challenge: messy data everywhere.what we did:
Standardized and cleaned datasets from multiple sources
Implemented a streamlined pipeline for continuous data quality checks
Used explainable ML models to help stakeholders trust the results
Standardized and cleaned datasets from multiple sources
Implemented a streamlined pipeline for continuous data quality checks
Used explainable ML models to help stakeholders trust the results
January 17, 2025 at 11:15 PM
During a project for a VP of Engineering, we faced a familiar challenge: messy data everywhere.what we did:
Standardized and cleaned datasets from multiple sources
Implemented a streamlined pipeline for continuous data quality checks
Used explainable ML models to help stakeholders trust the results
Standardized and cleaned datasets from multiple sources
Implemented a streamlined pipeline for continuous data quality checks
Used explainable ML models to help stakeholders trust the results
🔵 Phi 4: The 14B-parameter language model, which is an overachiever in the field of AI.
From the invention of Microsoft Research, Phi-4 reaches a new meaning of performance scaling: quality over brute force in size.
From the invention of Microsoft Research, Phi-4 reaches a new meaning of performance scaling: quality over brute force in size.
January 16, 2025 at 8:57 PM
🔵 Phi 4: The 14B-parameter language model, which is an overachiever in the field of AI.
From the invention of Microsoft Research, Phi-4 reaches a new meaning of performance scaling: quality over brute force in size.
From the invention of Microsoft Research, Phi-4 reaches a new meaning of performance scaling: quality over brute force in size.
🌐 Very recently, Google published a wonderful whitepaper on Generative AI Agents.
why it’s important?
While traditional LLMs generate texts only, they can't really plan, reason, or interact effectively with the world.
why it’s important?
While traditional LLMs generate texts only, they can't really plan, reason, or interact effectively with the world.
January 15, 2025 at 9:50 PM
🌐 Very recently, Google published a wonderful whitepaper on Generative AI Agents.
why it’s important?
While traditional LLMs generate texts only, they can't really plan, reason, or interact effectively with the world.
why it’s important?
While traditional LLMs generate texts only, they can't really plan, reason, or interact effectively with the world.
In a meeting, I was asked “What’s the first thing an ML consultant does?” Surprisingly, it’s not building models or writing code,it’s listening.
We dig into your challenges: What’s costing you time, money, or customers? Then we assess your data infrastructure to understand if it supports solutions.
We dig into your challenges: What’s costing you time, money, or customers? Then we assess your data infrastructure to understand if it supports solutions.
January 5, 2025 at 6:28 PM
In a meeting, I was asked “What’s the first thing an ML consultant does?” Surprisingly, it’s not building models or writing code,it’s listening.
We dig into your challenges: What’s costing you time, money, or customers? Then we assess your data infrastructure to understand if it supports solutions.
We dig into your challenges: What’s costing you time, money, or customers? Then we assess your data infrastructure to understand if it supports solutions.
What do I do as an ML/MLOps consultant?
Strategy First, Tools Second
I don’t dive into tools or code first, I start with your business goals.
Strategy First, Tools Second
I don’t dive into tools or code first, I start with your business goals.
December 29, 2024 at 11:05 PM
What do I do as an ML/MLOps consultant?
Strategy First, Tools Second
I don’t dive into tools or code first, I start with your business goals.
Strategy First, Tools Second
I don’t dive into tools or code first, I start with your business goals.
🌐Meta is shifting the AI paradigm from LLMs to LCMs.
Why it matters?
Instead of token-based predictions,Large Concept Models use “concepts" enabling them to:
🔹Plan outputs with explicit hierarchical structures
🔹Handle long contexts more efficiently
🔹Generalize across 200 languages without retraining
Why it matters?
Instead of token-based predictions,Large Concept Models use “concepts" enabling them to:
🔹Plan outputs with explicit hierarchical structures
🔹Handle long contexts more efficiently
🔹Generalize across 200 languages without retraining
December 27, 2024 at 8:37 PM
🌐Meta is shifting the AI paradigm from LLMs to LCMs.
Why it matters?
Instead of token-based predictions,Large Concept Models use “concepts" enabling them to:
🔹Plan outputs with explicit hierarchical structures
🔹Handle long contexts more efficiently
🔹Generalize across 200 languages without retraining
Why it matters?
Instead of token-based predictions,Large Concept Models use “concepts" enabling them to:
🔹Plan outputs with explicit hierarchical structures
🔹Handle long contexts more efficiently
🔹Generalize across 200 languages without retraining
Stats 101➡️ Train on all your data to understand relationships.
ML 101➡️ Split that data to test your model’s predictive muscle.
What works for stats doesn’t always fly in ML.
ML 101➡️ Split that data to test your model’s predictive muscle.
What works for stats doesn’t always fly in ML.
December 26, 2024 at 9:05 PM
Stats 101➡️ Train on all your data to understand relationships.
ML 101➡️ Split that data to test your model’s predictive muscle.
What works for stats doesn’t always fly in ML.
ML 101➡️ Split that data to test your model’s predictive muscle.
What works for stats doesn’t always fly in ML.
Top Trending Repo This Week:
Gemini API Cookbook!
A gateway to mastering multimodal AI with Google DeepMind’s latest tools.
github.com/google-gemin...
Gemini API Cookbook!
A gateway to mastering multimodal AI with Google DeepMind’s latest tools.
github.com/google-gemin...
GitHub - google-gemini/cookbook: Examples and guides for using the Gemini API
Examples and guides for using the Gemini API. Contribute to google-gemini/cookbook development by creating an account on GitHub.
github.com
December 21, 2024 at 7:45 PM
Top Trending Repo This Week:
Gemini API Cookbook!
A gateway to mastering multimodal AI with Google DeepMind’s latest tools.
github.com/google-gemin...
Gemini API Cookbook!
A gateway to mastering multimodal AI with Google DeepMind’s latest tools.
github.com/google-gemin...
Tool Overload?➡️ Welcome to the Chaos...
Kubernetes, Airflow, SageMaker, MLflow...the toolset for MLOps is vast. But:
⛔more tools ≠ better outcomes.
Kubernetes, Airflow, SageMaker, MLflow...the toolset for MLOps is vast. But:
⛔more tools ≠ better outcomes.
December 14, 2024 at 10:23 PM
Tool Overload?➡️ Welcome to the Chaos...
Kubernetes, Airflow, SageMaker, MLflow...the toolset for MLOps is vast. But:
⛔more tools ≠ better outcomes.
Kubernetes, Airflow, SageMaker, MLflow...the toolset for MLOps is vast. But:
⛔more tools ≠ better outcomes.
Choosing the right ML consultant isn’t easy.
Here’s a checklist every CTO/CEO should have:
✔️ Proven domain expertise: Do they know your industry? A healthcare ML model differs vastly from retail demand forecasting.
Here’s a checklist every CTO/CEO should have:
✔️ Proven domain expertise: Do they know your industry? A healthcare ML model differs vastly from retail demand forecasting.
December 11, 2024 at 10:36 PM
Choosing the right ML consultant isn’t easy.
Here’s a checklist every CTO/CEO should have:
✔️ Proven domain expertise: Do they know your industry? A healthcare ML model differs vastly from retail demand forecasting.
Here’s a checklist every CTO/CEO should have:
✔️ Proven domain expertise: Do they know your industry? A healthcare ML model differs vastly from retail demand forecasting.
AI + R&D = The Future of Innovation?
The Financial Times' "AI and the R&D Revolution" is a fascinating dive into how AI is reshaping R&D. Why does this matter? With $30 trillion in sales projected from new products over five years, the stakes couldn’t be higher:
www.ft.com/content/6480...
The Financial Times' "AI and the R&D Revolution" is a fascinating dive into how AI is reshaping R&D. Why does this matter? With $30 trillion in sales projected from new products over five years, the stakes couldn’t be higher:
www.ft.com/content/6480...
AI and the R&D revolution
How can this fast-changing technology help providers update products more efficiently and successfully at each stage of R&D
www.ft.com
December 6, 2024 at 10:10 PM
AI + R&D = The Future of Innovation?
The Financial Times' "AI and the R&D Revolution" is a fascinating dive into how AI is reshaping R&D. Why does this matter? With $30 trillion in sales projected from new products over five years, the stakes couldn’t be higher:
www.ft.com/content/6480...
The Financial Times' "AI and the R&D Revolution" is a fascinating dive into how AI is reshaping R&D. Why does this matter? With $30 trillion in sales projected from new products over five years, the stakes couldn’t be higher:
www.ft.com/content/6480...
Just read about how Uber mastered personalized marketing for 4 billion+ messages at scale!
Their challenge? Sending relevant recommendations without real-time user context like location or intent.
Their solution?👇🏻
Source: www.uber.com/en-GB/blog/p...
Their challenge? Sending relevant recommendations without real-time user context like location or intent.
Their solution?👇🏻
Source: www.uber.com/en-GB/blog/p...
Personalized Marketing at Scale: Uber’s Out-of-App Recommendation System
Out-of-app (OOA) communication (such as email, push, and SMS) is an important growth lever at Uber. It allows marketers, product owners, and operation teams to connect with users on a plethora of topics, including user promotions, new and favorite restaurants, etc. Building a system to personalize these communications presents unique and exciting challenges. In this blog post, we walk through these challenges and our journey in tackling them.
www.uber.com
December 5, 2024 at 8:12 PM
Just read about how Uber mastered personalized marketing for 4 billion+ messages at scale!
Their challenge? Sending relevant recommendations without real-time user context like location or intent.
Their solution?👇🏻
Source: www.uber.com/en-GB/blog/p...
Their challenge? Sending relevant recommendations without real-time user context like location or intent.
Their solution?👇🏻
Source: www.uber.com/en-GB/blog/p...
New to MLOps? Here's a beginner-friendly roadmap by Prasad Mahamulkar outlining the core steps:
Data ➡️ Training ➡️ Deployment ➡️ Monitoring.
Tools like DVC, MLflow, and Evidently AI make it seamless!
Data ➡️ Training ➡️ Deployment ➡️ Monitoring.
Tools like DVC, MLflow, and Evidently AI make it seamless!
December 4, 2024 at 8:54 PM
New to MLOps? Here's a beginner-friendly roadmap by Prasad Mahamulkar outlining the core steps:
Data ➡️ Training ➡️ Deployment ➡️ Monitoring.
Tools like DVC, MLflow, and Evidently AI make it seamless!
Data ➡️ Training ➡️ Deployment ➡️ Monitoring.
Tools like DVC, MLflow, and Evidently AI make it seamless!
a killer MLOps tip:
When starting a project, think pipelines not scripts.
Every MLflow pipeline has:
1️⃣ Code
2️⃣ Environment (conda/docker)
3️⃣ Execution logic (MLproject file)
Structure = Success.
When starting a project, think pipelines not scripts.
Every MLflow pipeline has:
1️⃣ Code
2️⃣ Environment (conda/docker)
3️⃣ Execution logic (MLproject file)
Structure = Success.
December 3, 2024 at 8:00 PM
a killer MLOps tip:
When starting a project, think pipelines not scripts.
Every MLflow pipeline has:
1️⃣ Code
2️⃣ Environment (conda/docker)
3️⃣ Execution logic (MLproject file)
Structure = Success.
When starting a project, think pipelines not scripts.
Every MLflow pipeline has:
1️⃣ Code
2️⃣ Environment (conda/docker)
3️⃣ Execution logic (MLproject file)
Structure = Success.
The future of MLOps? Automation.
I think:
📍Auto-retraining for model drift.
📍Feature stores for seamless data sharing.
📍Tools that even non-techies can use.
The easier MLOps gets, the faster ML will scale across industries.
I think:
📍Auto-retraining for model drift.
📍Feature stores for seamless data sharing.
📍Tools that even non-techies can use.
The easier MLOps gets, the faster ML will scale across industries.
November 30, 2024 at 7:02 PM
The future of MLOps? Automation.
I think:
📍Auto-retraining for model drift.
📍Feature stores for seamless data sharing.
📍Tools that even non-techies can use.
The easier MLOps gets, the faster ML will scale across industries.
I think:
📍Auto-retraining for model drift.
📍Feature stores for seamless data sharing.
📍Tools that even non-techies can use.
The easier MLOps gets, the faster ML will scale across industries.
Chip Huyen’s new book is coming:
🔴"AI Engineering: Building Applications with Foundation Models"🔴
Her first book on ML systems was a game-changer. This one dives into foundation models and building real-world AI apps.
@chiphuyen.bsky.social, you’ve done it again. 👍🏻
More thoughts soon!
🔴"AI Engineering: Building Applications with Foundation Models"🔴
Her first book on ML systems was a game-changer. This one dives into foundation models and building real-world AI apps.
@chiphuyen.bsky.social, you’ve done it again. 👍🏻
More thoughts soon!
November 29, 2024 at 7:27 PM
Chip Huyen’s new book is coming:
🔴"AI Engineering: Building Applications with Foundation Models"🔴
Her first book on ML systems was a game-changer. This one dives into foundation models and building real-world AI apps.
@chiphuyen.bsky.social, you’ve done it again. 👍🏻
More thoughts soon!
🔴"AI Engineering: Building Applications with Foundation Models"🔴
Her first book on ML systems was a game-changer. This one dives into foundation models and building real-world AI apps.
@chiphuyen.bsky.social, you’ve done it again. 👍🏻
More thoughts soon!
Feature engineering is where I found my creative groove. On one project, I had two features that didn’t make much sense on their own, but together, they told a story.
November 28, 2024 at 8:20 PM
Feature engineering is where I found my creative groove. On one project, I had two features that didn’t make much sense on their own, but together, they told a story.
Imagine cutting customer issue resolution time by nearly 30%!
@LinkedIn just did it with AI-powered knowledge graphs!
By integrating retrieval-augmented generation (RAG) with knowledge graphs (KGs), LinkedIn transformed customer service.
#AICaseStudy👇🏻
@LinkedIn just did it with AI-powered knowledge graphs!
By integrating retrieval-augmented generation (RAG) with knowledge graphs (KGs), LinkedIn transformed customer service.
#AICaseStudy👇🏻
November 27, 2024 at 8:37 PM
Imagine cutting customer issue resolution time by nearly 30%!
@LinkedIn just did it with AI-powered knowledge graphs!
By integrating retrieval-augmented generation (RAG) with knowledge graphs (KGs), LinkedIn transformed customer service.
#AICaseStudy👇🏻
@LinkedIn just did it with AI-powered knowledge graphs!
By integrating retrieval-augmented generation (RAG) with knowledge graphs (KGs), LinkedIn transformed customer service.
#AICaseStudy👇🏻
Multimodal Safety Revolution?
New paper by GenAI at Meta suggests that Llama Guard 3 Vision, the latest safeguard for human-AI multimodal interactions, excels in detecting harmful text and image prompts/responses.
New paper by GenAI at Meta suggests that Llama Guard 3 Vision, the latest safeguard for human-AI multimodal interactions, excels in detecting harmful text and image prompts/responses.
November 25, 2024 at 7:57 PM
Multimodal Safety Revolution?
New paper by GenAI at Meta suggests that Llama Guard 3 Vision, the latest safeguard for human-AI multimodal interactions, excels in detecting harmful text and image prompts/responses.
New paper by GenAI at Meta suggests that Llama Guard 3 Vision, the latest safeguard for human-AI multimodal interactions, excels in detecting harmful text and image prompts/responses.
Asking me the hardest part of MLOps? Alignment.
I’ve seen MLOps initiatives fail, not because the tech didn’t work, but because teams weren’t aligned:
I’ve seen MLOps initiatives fail, not because the tech didn’t work, but because teams weren’t aligned:
November 24, 2024 at 5:01 PM
Asking me the hardest part of MLOps? Alignment.
I’ve seen MLOps initiatives fail, not because the tech didn’t work, but because teams weren’t aligned:
I’ve seen MLOps initiatives fail, not because the tech didn’t work, but because teams weren’t aligned:
Curious about how MLOps evolves for 6G? This article introduces tailored pipelines:
🔹 'RLOps' for adaptive RL,
🔹 'FedOps' for biased federated learning,
🔹 and 'GenOps' for energy-efficient generative AI.
A must-read for AI-native 6G insight
👇 arxiv.org/abs/2410.18793
🔹 'RLOps' for adaptive RL,
🔹 'FedOps' for biased federated learning,
🔹 and 'GenOps' for energy-efficient generative AI.
A must-read for AI-native 6G insight
👇 arxiv.org/abs/2410.18793
November 23, 2024 at 7:39 PM
Curious about how MLOps evolves for 6G? This article introduces tailored pipelines:
🔹 'RLOps' for adaptive RL,
🔹 'FedOps' for biased federated learning,
🔹 and 'GenOps' for energy-efficient generative AI.
A must-read for AI-native 6G insight
👇 arxiv.org/abs/2410.18793
🔹 'RLOps' for adaptive RL,
🔹 'FedOps' for biased federated learning,
🔹 and 'GenOps' for energy-efficient generative AI.
A must-read for AI-native 6G insight
👇 arxiv.org/abs/2410.18793
Reposted by Amir Zeinali
If you are looking for LLM stuff to read/study, I added lots of bonus from-scratch coding resources over the last few months (from implementing Llama 3.2 to preference tuning with DPO): github.com/rasbt/LLMs-f...
I hope you find them useful!
PS: more to come soon when I am back from a trip!
I hope you find them useful!
PS: more to come soon when I am back from a trip!
https://github.com/rasbt/LLMs-fro…
November 20, 2024 at 11:46 PM
If you are looking for LLM stuff to read/study, I added lots of bonus from-scratch coding resources over the last few months (from implementing Llama 3.2 to preference tuning with DPO): github.com/rasbt/LLMs-f...
I hope you find them useful!
PS: more to come soon when I am back from a trip!
I hope you find them useful!
PS: more to come soon when I am back from a trip!