Bhavin Moriya, Ph.D’s Post

Just built a side-by-side semantic search playground Python for real transformer embeddings, Rust for raw high-throughput cosine similarity on 10k vectors. Python side: sentence-transformers + NumPy, loads actual models, optional SVD compression, gives meaningful top-k results instantly. Rust side: ndarray doing full 10k × 384 matrix ops in release mode synthetic but stupidly fast, zero surprises. Startup difference is night and day: Python waits for model download and weights, Rust just flies from first cargo run. Memory profile tells the story Python carries the full ML stack, Rust stays lean and predictable even at scale. Same cosine math, same top-k goal, yet Rust’s memory safety and zero-cost abstractions make the benchmark feel like cheating. The gap isn’t hype it’s observable the moment you switch from prototype to production-grade retrieval. If you’re already writing Python or JS for vector search and wondering where the latency and safety edge actually lives… this repo is the signal. Drop a comment or DM if you are building with Rust. #Rust #AI #ML

𝐔𝐧𝐝𝐞𝐫𝐬𝐭𝐚𝐧𝐝𝐢𝐧𝐠 𝐜𝐨𝐫𝐫𝐞𝐥𝐚𝐭𝐢𝐨𝐧𝐬 𝐦𝐚𝐝𝐞 𝐝𝐚𝐭𝐚 𝐚𝐧𝐚𝐥𝐲𝐬𝐢𝐬 𝐦𝐨𝐫𝐞 𝐢𝐧𝐭𝐞𝐫𝐞𝐬𝐭𝐢𝐧𝐠 𝐟𝐨𝐫 𝐦𝐞 While exploring datasets in Python recently, I spent some time understanding how correlation works between variables. Using pandas, it’s surprisingly easy to calculate a correlation matrix and see how different columns relate to each other. Sometimes two variables move together strongly, and sometimes there’s almost no relationship at all. What I found interesting is that correlations can quickly highlight patterns that might not be obvious just by looking at raw numbers. Still learning how to interpret these relationships properly, but it’s definitely making the analysis process more insightful. #Python #Pandas #DataAnalytics

🚀 Hook: I started building my first interactive data dashboard using Python… and here’s what I’ve learned so far 👇 --- 💡 Caption: After working on my EDA tool, I decided to level up my skills by building a data dashboard. Right now, I’m in the process of building it using: - Python - Streamlit - Plotly So far, I’ve learned: ✅ How to load and clean data ✅ How to create basic charts ✅ How to structure a simple dashboard layout Still facing some issues while running the app — but solving them step by step 💪 This journey is teaching me one important thing: 👉 You don’t need to be perfect to start… you just need to start. --- 💬 If you’ve built dashboards before, any tips would be helpful! 👇 Follow me to see the final version soon. --- 🔥 Hashtags: #DataAnalytics #Python #LearningInPublic #Streamlit #Plotly #BeginnerJourney #BuildInPublic #Tech #AI #Projects

I built a machine learning web app that predicts whether a loan will be approved or rejected based on applicant financial data.In this project, I used Python, Scikit-learn, and Streamlit. I trained multiple models including Naive Bayes, KNN, and Logistic Regression, and selected the best-performing model for final deployment. Link:-https://lnkd.in/giKaMpyz

We are taking the training wheels off. 🚲 In Part 7, we used the "Easy Button" to build an AI agent. Today, in Part 8, we are opening up a Jupyter Notebook and building a custom RAG pipeline from absolute scratch using Python. If you want to move from "Full-Stack Developer" to "Data Scientist / AI Architect," you have to understand the math beneath the magic. In this tutorial we cover: 🔪 Programmatic Text Chunking 🔢 Generating Vector Embeddings (text-embedding-004) 📐 Calculating Cosine Similarity with numpy to build a semantic search engine. Read the full tutorial here: https://lnkd.in/ewtWxBT6 #Python #DataScience #MachineLearning #VertexAI #GoogleCloud #VectorSearch

If you're struggling to get started with LangGraph, I built a small text quality checker API that covers parallel branches, conditional routing, retry loops, and state management. All in one place. Blog: https://lnkd.in/gXwuUnvT #LangGraph #Python #AI

Recently, I implemented both Linear Regression and Logistic Regression from scratch using Python and NumPy, emphasizing vectorization and mathematical understanding. For both models, I utilized two feature vectors and created animated visualizations of the learning process, illustrating how the decision boundary (for Logistic Regression) and regression plane/line (for Linear Regression) evolve step by step during gradient descent. A major goal was to ensure the code was scalable, efficient, and mathematically transparent. Key aspects of the implementation include: • Fully vectorized code, avoiding unnecessary loops • Significant speed improvement due to vectorization, enhancing training efficiency • Generalization capability from 2 features to n-feature input spaces • Structured for straightforward future implementation of Lasso (L1) and Ridge (L2) regularization • Visualization currently limited to 2D feature space for interpretability Building these models from scratch deepened my understanding of the underlying mathematics, particularly gradient descent, cost functions, normalization, decision boundaries, and parameter updates. Writing the algorithm myself proved to be a more insightful learning experience than simply using a library. Code can be found at: https://lnkd.in/ghqdDxMg #MachineLearning #LinearRegression #LogisticRegression #LearningByBuilding

Claude just diagnosed me with a classic developer bug 😂 After hours of learning Python — functions, loops, dictionaries, if/else, and AI agent architecture — I started asking the same questions twice. Claude's response? ``` while awake == True:   ask_questions()       if questions == repeat:     print("Go to sleep Anil! 😄")     break ``` Turns out even humans need a break statement. 😄 The grind is real. But so is the progress. 💪 #Python #AI #MachineLearning #CareerChange #AIAgent #LearningToCode #Claude #100DaysOfCode

Today, I focused on working with NumPy arrays. Building a solid foundation for data manipulation and analysis. Here’s what I practiced: 🔹 Created a 1D array with values from 1 to 15 🔹 Built a 2D array (3×4) filled with ones 🔹 Generated a 3×3 identity matrix 🔹 Explored key array properties like shape, type, and dimensions 🔹 Converted a regular Python list into a NumPy array This session helped me better understand how data is structured and handled in numerical computing. Getting comfortable with arrays is definitely a crucial step toward more advanced data analysis and machine learning tasks. Looking forward to building on this momentum 💡 #AI #MachineLearning #Python #NumPy #DataAnalysis #M4ACE

🚀 Unlocking the Power of Numerical Python with NumPy! I just finished a deep dive into NumPy, the foundational package for numerical computation in Python. It’s incredible how much complexity you can simplify with just a few lines of code! Here’s a quick recap of the core concepts I explored: Array Creation: Effortlessly generating data using np.zeros(), np.ones(), np.arange(), and np.linspace(). I also tapped into np.random.random() for statistical simulations. Indexing & Slicing: Mastering access to specific elements and rows. Boolean indexing (e.g., a[a > 2]) is a total game-changer for filtering data quickly. Mathematical Operations: Performing lightning-fast element-wise operations and using built-in functions like np.sqrt() for efficient transformations. Statistical Analysis: Calculating mean, median, and std across different axes. I especially appreciated learning about np.nanmean to handle missing values without breaking the code. Data Cleaning: Putting it all together to identify and remove extreme values (outliers) from a dataset to ensure cleaner, more accurate analysis. NumPy is an indispensable tool for Data Science, Machine Learning, and Scientific Computing. Its efficiency makes it a "must-have" in any Python developer's toolkit. #Python #NumPy #DataScience #MachineLearning #Coding #DataAnalysis #ProgrammingTips