Romee Panchal retweeted
A useful rule in AI engineering:
Don’t start with the model. Start with the failure mode.
Ask:
• What can go wrong?
• Can rules solve part of it?
• Where does ML add value?
• How will we know it works?
Good AI starts with good problem framing.
Romee Panchal retweeted
The first real evidence that the days of LLM Scaling laws are over.
Introducing SCOPE: the world's most efficient Neural Planner. 🔍📊
We tested SCOPE vs Frontier LLMs for planning tasks on TextCraft (text version of Minecraft) and here are the results: ⤵️
- SCOPE Runs 55x faster than GPT 3.5 (3 seconds vs 164 seconds)
- SCOPE is 160,000 smaller than GPT 4o (11M parameters vs 1.8T parameters)
- SCOPE is more accurate on Planning tasks (56%) than frontier LLM models
The age of efficient AI models starts now.
🔗📌 Read the full write up here: skyfall.ai/blog/scope-hierar…
Romee Panchal retweeted
ML Engineering:
Learn programming -> machine learning concepts -> system design -> deployment strategies -> cloud computing -> focus on practical projects.
Romee Panchal retweeted
If you already know Python and the basics of neural networks, you are ahead of many. These fundamentals will help you quickly get up to speed with Gen AI.
To better understand NNs and language models, you can go through Karpathy's Zero To Hero series. Then, start by training a small transformer-based model using the Hugging Face transformers package on a toy dataset. This will help you understand model training, tokenization, and inference.
After this learn how to fine-tune a Pre-trained LLM for a specific task. Then, you can move on to learning about RAGs and agentic AI systems.
You can use ChatGPT or Claude for project ideas.
This is more of a general guide to making your way up to Gen AI, not a definitive one. You can set a target to build some big project and then use back-chaining to know about what you need to learn to do that project. Then build your knowledge/skills from there.
Romee Panchal retweeted
You can understand machine learning or deep learning algorithms from two different aspects. Both aspect answer a different thing.
Code: how does it work?
Math: why does it work?
I think both aspects are important.
Romee Panchal retweeted
Linear Regression implementation from scratch in numpy:
colab.research.google.com/dr…
Romee Panchal retweeted
Replying to @DSDecoded
One needs to figure out what they want to achieve and then balance their learning accordingly. If you want to go into research, then you definitely need advanced maths. However, if you are going into DS/MLE roles, more practical and engineering skills are needed.
Romee Panchal retweeted
🚀 DeepSeek-R1 is here!
⚡ Performance on par with OpenAI-o1
📖 Fully open-source model & technical report
🏆 MIT licensed: Distill & commercialize freely!
🌐 Website & API are live now! Try DeepThink at chat.deepseek.com today!
🐋 1/n
Romee Panchal retweeted
Neural network implementation from scratch in Python with Numpy.
[full implementation link in comments]
Romee Panchal retweeted
Replying to @adheeeep @Blacklazer77
Some ideas for ML projects:
Basic Projects: Housing Price Predictor, Image Classification with MNIST, Sentiment Analysis for Movie Reviews, Customer Churn Prediction, and Music Genre Classification.
Advanced Projects: Language Translation Model, Image Caption Generation, Chatbot for Customer Support using RNN or Transformers, Building a GAN to Generate New Faces, Link Prediction in Knowledge Graphs
Romee Panchal retweeted
Replying to @Abhishekcur
For machine learning-specific reasons, check out the following playlist. It is really good.
Romee Panchal retweeted
What really works when learning ML?
- Read concepts from one good source. Don’t hop from one resource to another.
- Prioritize solving problems. This step will help reinforcing the concepts and identify gaps.
- Implement ML algorithms from scratch from your memory. Don’t look up details of an algorithm while implementing it. This is crucial to put things into your long-term memory.
- Stay curious. Ask why a particular technique is used for a solution. That will make you a better engineer.
Romee Panchal retweeted
Replying to @minamisatokun @3blue1brown
Simply watching a video gives us a feeling of understanding. We don't fully know whether we truly know a topic. I think to effectively revise concepts and test your knowledge, ask Claude or ChatGPT to generate a list of questions on a topic and solve them. Then, ask the LLM to provide step-by-step solutions to each question. Finally, compare your solutions with AI. You can also use a question bank as well for this purpose.
Romee Panchal retweeted
Practical tips to build mathematical intuition:
- Focus on Understanding "Why"
- Practice Active Problem-Solving
- Build From Fundamentals
- Engage in Real Applications
- Challenge yourself with progressively harder problems.
- Revisit concepts from different angles
Romee Panchal retweeted
If you want to get work done, start doing it. That's the only way. No matter how many productivity videos you watch, how many articles you read, or how many different task-scheduling systems you try, they won't help.
If the work seems overwhelming, break it down into smaller chunks and attempt them one at a time. But still, it is you who has to carry out the work. It is you who has to take the action. So, do it.
Romee Panchal retweeted
Which ML algorithm do people generally use for tabular data?
Baselining:
Linear/logistic regression
kNN
Main model:
XGBoost
Romee Panchal retweeted
Why do want to learn machine learning? Is it because everyone talks about it? Or, are you deeply interested in learning this skill seriously? Answer this question before jumping the ship.
Romee Panchal retweeted
Roadmap to learn machine learning.
Step 1: Read about a machine learning algorithm (use any popular source, don't put too much time into deciding on the best source).
Step 2: Implement it in Python from scratch (this should give you a much better understanding of the inner workings of the algorithm).
Step 3: Compare your implementation against established libraries such as scikit-learn, and PyTorch (here you will learn about any potential issues, learn the significance of each hyperparameter, and improve your coding skills for ML).
Step 4: Pick a few datasets and solve the underlying ML challenges. Better if you choose from famous datasets from Kaggle (learn to analyse data, create features, and build models in the process., evaluate the performance, go through others' solutions (that's why Kaggle) and repeat. You will learn some serious ML skills).
Step 5: Repeat the step 1-4 with other algorithms.
Additional Step: Learn about model deployment, productionization, and monitoring.
The steps assume that you have the following prerequisites:
1. Python
2. Fundamental mathematics (Linear Algebra, Calculus)
3. Basic statistics and probability
Romee Panchal retweeted
Replying to @yoobinray
Downtime is super important for the mind to recharge and organise information.









