Posts

Showing posts with the label Data

This Week I Learned - Week 29 2026

Image
This Week I Learned -  * Microsoft's Phi-4 Mini is widely regarded as one of the most capable models under 4 billion parameters. It punches significantly above its weight class, outperforming older 7B and even 70B models in specific reasoning tasks. It runs at a smooth 13–18 tokens per second on flagship devices with 8 GB+ RAM (like the iPhone 16 Pro/17 Pro or Galaxy S25/S26 series). * From The Batch : - PrismML, a Caltech spinout backed by Khosla Ventures, publicly released compressed versions of Alibaba’s Qwen model on Tuesday, shrinking it from 54 GB to under 4 GB so all 27 billion parameters run on an iPhone (15 or newer). PrismML achieves the compression by reducing the models to ternary or binary quantization, reducing each value from 16 bits to just two or three possible values.   - Two people asking the same model for business feedback, one in Hindi and one in Russian, may receive substantively different advice not because they’re using different models, but because th...

HOW TO Download Open Food Facts Data for Any Country Using Google Colab and Push It to GitHub

Image
Open Food Facts has data on 4.6+ million food products worldwide. There are multiple ways to access that data but the simplest way to get a subset of the data with chosen parameters is to  download it as a CSV file  through the website's search interface. But if you need entire data for one country, you can filter the  7.64 GB file  (size at the time of writing) hosted on Hugging Face as there are no natively available exports dedicated to specific countries currently . Hugging Face acts as a CDN for large ML datasets — it not only stores the full Parquet file but also handles versioning, caching, and fast global downloads. This guide co-written with Meta AI shows you how to use Google Colab to filter the dataset for just one country, add direct product links, and save the result to a public GitHub repo. Takes < 30 minutes total based on your familiarity with these tools & file size. No software installs needed. What you’ll end up with: A CSV like india_pr...

This Week I Learned - Week 9 2026

Image
This Week I Learned -  *  Supabase is an open-source Backend-as-a-Service (BaaS) built around PostgreSQL. It is a Firebase alternative that can help you set up a backend in less than 2 minutes. * Add the possibility of an Internet Service Provider-level blocking of a service to the list of things you should debug when your code isn't working. Now if your perfectly working code accesses some paid premium AI-Model APIs suddenly throws errors, you'll be losing tokens while you figure that it was in fact your ISP that was glitching.  A subset of users based in India accessing Supabase over JioFiber, Reliance Jio's fiber-to-the-home broadband service, faced routing/block issues. Supabase’s own infrastructure was fully operational but users on JioFiber (and some other Indian ISPs) couldn’t connect to Supabase services like Dashboard, Auth, Edge Functions, Postgres in ap-south-1, etc. The root cause turned out to be ISP-level DNS resolution failure. The ISP’s DNS serv...

Data Insights: China is the top import partner for most countries in the world

Image
China is the largest source of merchandise goods (by value) for around two-thirds of countries worldwide .  In many countries, China has overtaken the United States as the largest origin of their imported goods.  China’s trade surplus has officially topped $1 trillion, marking the first time any country has ever reached such an excess of exports over imports. 🌏 Our World in Data is a project of Global Change Data Lab, a nonprofit based in the UK.  Our World in Data offers a curated collection of charts on its website as well as datasets, indicators and  Grapher Chart API .  Our World in Data (OWID) built their own open source data visualization platform Grapher because Datawrapper, Tableau, Plotly, various libraries based on d3 were not suitable.

This Week I Learned - Week 46 2025

Image
* Every AI application startup is likely to be crushed by rapid expansion of the foundational model providers. The foundational provider introduces continual chaos into the entire ecosystem at a rate never before seen, to a degree such that downstream providers can never get established.  It’s not a one-time sea change, it’s continual tsunamis. There are two ways AI application startup founders can make money: - Make a flash-in-the-pan app that generates a ton of cash and bank the cash (my estimate is that you have about 12-18 months cashflow generation) - Make a good enough app that you get acquired by one of the big players for sufficient equity Sea changes are now happening on a 9-12 month cycle. Very few startups can turn into a mature business in that timeframe - and by mature, I mean having all the boring stuff like sales relationships and brand recognition. The physical moat is the only one that's like "large rocks that can offer cover from the continually crashing wave...

This Week I Learned - Week 43 2025

Image
This Week I Learned -  * AWS supports more than a third of the 100,000 busiest websites in the world. * AWS experienced a significant global outage that disrupted various online platforms, including social media, gaming, and financial services for about 8 hours. The company traced the issue to a DNS resolution problem in the US-EAST-1 Region, caused by an error in Amazon's EC2 internal network. This affected services like DynamoDB, SQS, Amazon Connect, and others. According to Amazon’s service status page, 37 AWS services were impacted, including prominent tools like AWS Lambda, Amazon CloudFront, Amazon EC2, AWS Systems Manager, Amazon SageMaker, Amazon API Gateway, AWS Security Token Service, Amazon Connect, and Amazon VPC Lattice. * US-EAST-1 location in northern Virginia is the oldest and largest site for AWS. The site suffered outages in 2021 and 2020. * The AWS outage disrupted several online services for hours, affecting platforms like Amazon, Alexa, Duolingo, Snapchat, Fort...

HOW TO View Large CSV File From A GitHub Repo Without Downloading It Locally

Image
Here are quite a few ways to view large CSVs from a GitHub repo without downloading them locally -  1) Flat Viewer is a free online viewer for CSV, JSON, and other tabular files stored in public GitHub repos — it displays them in a sortable, filterable table without you having to download them. The FlatGithub URL format is: https://flatgithub.com/OWNER/REPO?filename=PATH_TO_FILE If your file is: https://github.com/ mvark/indiafoodstats/blob/main/Brands/Amul.csv The FlatGithub link will be: https://flatgithub.com/ mvark/indiafoodstats/blob/main/Brands/Amul.csv What you can do with Flat Viewer - View the CSV as a table (no GitHub size preview limit) Sort columns Search in the dataset Download the filtered view as CSV or JSON 2) GitHub + Raw + CSV Explorer Go to your CSV in GitHub Click Raw Copy the URL (it will look like https://raw.githubusercontent.com/.../file.csv) Go to CSV Explorer  (requires sign up) Import the data using Raw CSV file  3)  Through your own D...

Student Perks: Free Software, Certs, and Learning Bundles

Image
There are many free and discounted certifications, software bundles, tools, services, and learning resources available for students and some for the general public too (marked with asterisk) - Oracle University is offering 5 free certification exam attempts * as part of its Race to Certification Challenge 2025 that will run till October 2025 and 3 always-free Foundation course certifications *. GitHub Student Developer Pack : Github offers Copilot Pro for free, 180 monthly Codespaces hours for cloud coding, unlimited private repositories with GitHub Pro or Team, and dozens of premium tools in the Student Developer Pack. Below is the list with a link to the dedicated landing page (where available) for that tool's GitHub student offer - Developer Tools & IDEs GitHub Copilot Student Offer JetBrains Student Pack Appwrite Student Education Plan GitLens Student Pack GitKraken Student Pack GitHub Desktop Polypane GitHub Student Bootstrap Studio Student Pack ...

Oracle University's Race to Certification 2025

Image
Race to Certification 2025 learning and certification challenge running from July 1 to October 31, 2025 offers an opportunity to earn free professional certification exam attempts for the following exam titles: AI   Oracle Cloud Infrastructure 2025 AI Foundations Associate | 1Z0-1122-25 Oracle Cloud Infrastructure 2025 Data Science Professional | 1Z0-1110-25 Oracle AI Vector Search Professional | 1Z0-184-25 Oracle Cloud Infrastructure 2025 Generative AI Professional | 1Z0-1127-25 OCI  Oracle Cloud Infrastructure 2025 Foundations Associate | 1Z0-1085-25 Orac...

This Week I Learned - Week #26 2025

Image
This Week I Learned -  * Gemini CLI is an open-source AI agent that brings Gemini directly into your terminal, with MCP support for extensibility and Human in the Loop for oversight.  * Prompting only changes how the model responds. Fine-tuning changes the weights—but it’s costly, brittle, and still static. Neither gives the model new knowledge in real-time. With Retrieval-Augmented Generation (RAG) , you “Retrieve” the most relevant data for the user’s query from your database, use this data to “Augment” the prompt you send to the LLM, and then let it “Generate” a response based on the user query + prompt + retrieved data. RAG doesn’t modify the model; it modifies the input. It retrieves relevant context (docs, tickets, policies) and feeds it into the prompt at inference time. No retraining. Just better answers, grounded in your own data. *  Google Apps Script Web Apps can act as powerful servers for open protocols like Model Context Protocol (MCP) and Agent2Agent (A2A)...

This Week I Learned - Week #25 2025

This Week I Learned -  * ScyllaDB Team's Database Benchmarking for Performance Masterclass - Best Practices for Running Your Own Tests *  OpenRefine is an open source tool for working with messy data: cleaning it; transforming it from one format into another; and extending it with web services and external data. * A.I. text generation — which is what chatbots do — used 10 times as much energy compared with simple classification tasks like sorting emails into folders. People looking for facts are better off just using a search engine, since generative A.I. can “hallucinate” false information. Longer answers and those that use a reasoning process generate more emissions. - NYT *  There are three major areas where humans either are, or will soon be, more necessary than ever: trust, integration and taste . - NYT *  Y Combinator AI Startup School Highlights When asked about qualities he looks for in people, Satya Nadella outlined three non-negotiables: Clarity builders...