Get live statistics and analysis of Aarno's profile on X / Twitter

You still use pip and requirements.txt ?

If you’re getting into PyTorch, give this a read. It discusses the usability, design patterns and implementation ideas behind the framework. A few bits and pieces that can help you build a good foundation.

As a data scientist, never count Software Engineering out of your domain. Get very very comfortable with - 1. Docker 2. Redis 3. Postgres 4. Qdrant 5. Git / Github / GitHub Actions No matter what part of AI you work in, these will always go a very long way. A good setup will always set you a class apart from your peers.

Most ML courses stop at regression, classification, and clustering. But real world data is messy and often takes extensive effort. Here are 3 underrated models you should know for anomaly detection: > Local Outlier Factor: Detects anomalies based on local density > Isolation Forest: Isolates rare points instead of profiling normal ones > RRCF (Robust Random Cut Forest): Amazon’s scalable model for streaming anomaly detection. Perfect for fraud detection, sensor drift, and log monitoring. Implement these; they’ll make you better than 90% of “ML engineers.”

Learn MLOps by implementing this pipeline. This is a great repo which covers every stage of the pipeline, introducing you to some essential principles. This can be covered in 10 weeks. You learn > Project setup, Model Monitoring > Configuration Management > Data Version Control > Model packaging and deployment > Prediction Monitoring Along the way, learn the base toolset needed to setup an MLOps pipeline. Link in the comments

There will never be a better time to start with Reinforcement Learning. If you’re already into ML, this course should be the place to start with. Other options > Stanford CS234 > Barton & Sutto (The OG) > Explore OpenAI Gym & Spinning Up RL is going mainstream !!

If you’re looking to get into deep learning; there are a number of quality courses you can find > Stanford CS230 - Andrew NG (A new version is out) > MIT S191 (updated almost every year) > UC Berkeley CS182 > IITM CS6910 (NLP route into DL) Here’s one book that I tried and loved. Grokking DL is a seriously awesome intro to Deep learning. It sets up intuition well, uses great examples and leaves you with a good understanding about Deep Learning before you venture deeper. Give it a go !

If you’re going to apply for ML Engineering roles, specifically working in LLM based applications, there’s a few skills that often go overlooked despite being quite obvious > Cost optimisation > Latency & Throughput, TTFT > Data pipelines > Batch inferencing > Evaluation pipelines > Model safety & guardrails Implementing an end to end project that showcases at least two of the above listed skills can greatly boost your chances of clearing the interview. A great idea is only worth its execution :)

System Design, particularly in ML can have a varying number of requirements- but the ideas generally fall into a number of categories. Depending upon context, specifications change, but baseline components remain unchanged. Watch this video to understand how these interviews are tackled. Great strategy and execution shared.

Most classical ML books only talk about learning parameters. They lack > inference algorithms (MC-MC, EM) > Neural Networks as probabilistic function approximations > Bayesian DL, VAEs > ML models in PGM terms Kevin Murphy’s Probabilistic ML bridges these gaps. It talks about a unified probabilistic framework and what “learning from data” actually means from a Bayesian perspective. It’s an advanced book but definitely a must read. Changes your world.

We’re starting with this baby today. A lot of topics covered in this, I have, at some point, already worked with. But this is a great formalisation of concepts around LLM based apps and infra > Data Engineering > Ops > Fine tuning (SFT, RL) > Evals > Inference Optimisation Plan is to derive at least 2 end to end projects out of this.

If you're headed for Applied AI roles today, the best of projects will fail to impress unless you set up exhaustive evaluations. The real value comes from these metrics - which ultimately run the business you wanna work for. A good project + evaluations pipeline (real time / batch ) will go far. Follow this blog to learn some of the most import metrics - and how / where to use them.

One Shot. One Opportunity. 🙏🏻 Cook or get cooked.

You still use pip and requirements.txt ?

One remains. Rest gotta go. Your choice ?

As a data scientist, never count Software Engineering out of your domain. Get very very comfortable with - 1. Docker 2. Redis 3. Postgres 4. Qdrant 5. Git / Github / GitHub Actions No matter what part of AI you work in, these will always go a very long way. A good setup will always set you a class apart from your peers.

October was a great month for me > started posting here (still need to do more), got 1k followers > multiple startups reached out > got a great bunch of people on this platform > resigned from my current job (career update soon🥰) > learning and building more actively on a personal level now You can just do things. Follow me to see where it goes, reach out because I’m always happy to help. Big stuff coming.

If you’re getting into PyTorch, give this a read. It discusses the usability, design patterns and implementation ideas behind the framework. A few bits and pieces that can help you build a good foundation.

Learn MLOps by implementing this pipeline. This is a great repo which covers every stage of the pipeline, introducing you to some essential principles. This can be covered in 10 weeks. You learn > Project setup, Model Monitoring > Configuration Management > Data Version Control > Model packaging and deployment > Prediction Monitoring Along the way, learn the base toolset needed to setup an MLOps pipeline. Link in the comments

If you’re going to apply for ML Engineering roles, specifically working in LLM based applications, there’s a few skills that often go overlooked despite being quite obvious > Cost optimisation > Latency & Throughput, TTFT > Data pipelines > Batch inferencing > Evaluation pipelines > Model safety & guardrails Implementing an end to end project that showcases at least two of the above listed skills can greatly boost your chances of clearing the interview. A great idea is only worth its execution :)

The Grind continues. We’re so back.

Balloon day 🙏🏻 Also a month since I joined this community. Thought I’d post for a day or two and leave. But something got me stuck here and wanting to open up and share more. Thank you to everyone who followed. 🫡 Gonna keep building up!

Most ML courses stop at regression, classification, and clustering. But real world data is messy and often takes extensive effort. Here are 3 underrated models you should know for anomaly detection: > Local Outlier Factor: Detects anomalies based on local density > Isolation Forest: Isolates rare points instead of profiling normal ones > RRCF (Robust Random Cut Forest): Amazon’s scalable model for streaming anomaly detection. Perfect for fraud detection, sensor drift, and log monitoring. Implement these; they’ll make you better than 90% of “ML engineers.”

There will never be a better time to start with Reinforcement Learning. If you’re already into ML, this course should be the place to start with. Other options > Stanford CS234 > Barton & Sutto (The OG) > Explore OpenAI Gym & Spinning Up RL is going mainstream !!

We’re starting with this baby today. A lot of topics covered in this, I have, at some point, already worked with. But this is a great formalisation of concepts around LLM based apps and infra > Data Engineering > Ops > Fine tuning (SFT, RL) > Evals > Inference Optimisation Plan is to derive at least 2 end to end projects out of this.

BTW, I use arch (well, kind of) Moved to EndeavourOS today. It’s based on Arch linux, or rather, it is Arch Linux but without the hassle of a complete manual setup. It comes bundled with Calamares installer, which helps set you up and get running much quicker. If you’re looking to move deeper into the linux arena, but cannot spend a lot of time learning and setting up from scratch, this is the way to go. Will be messing around with stuff this week, drop some suggestions on what I should try out as a first time arch user !!

Most classical ML books only talk about learning parameters. They lack > inference algorithms (MC-MC, EM) > Neural Networks as probabilistic function approximations > Bayesian DL, VAEs > ML models in PGM terms Kevin Murphy’s Probabilistic ML bridges these gaps. It talks about a unified probabilistic framework and what “learning from data” actually means from a Bayesian perspective. It’s an advanced book but definitely a must read. Changes your world.

An Awesome collection of MLOps tools compiled in a single repo. With infra, there's often a struggle to pick out the best tools for the job considering the number of possibilities there are. See this list to find the best ones for you.