FAQ

It's the personal blog of Perivitta Rajendran, an AI Engineer based in Kuala Lumpur, Malaysia. The site shares articles, guides, and lessons on machine learning and artificial intelligence — written to make complex AI topics simple, practical, and accessible.

Perivitta ("Peri") is an AI engineer who has worked in the field since 2021. She writes about building reliable, real-world AI systems, with a focus on practical outcomes rather than chasing perfect metrics.

Her main areas of focus are designing AI systems that combine language models with tools and data, building retrieval-based systems that help AI give grounded answers, building AI agents that can reason and use tools, and evaluating AI quality in real-world use.

She holds 13 certifications across three major cloud providers — Amazon Web Services (AWS), Google Cloud (GCP), and Microsoft Azure — covering cloud, machine learning, and data topics.

Through the Contact page, by email, on LinkedIn, or on GitHub. The Contact page also includes a simple message form (name, email, optional phone number, and your message).

She's based in Kuala Lumpur, Malaysia, and is open to conversations, collaborations, and interesting problems in AI and data.

Home, About, Posts (the blog), Courses (structured learning), Explore, and Contact, all in the navigation bar at the top of every page. Explore opens a larger menu organized into three groups: Model Rankings (the AI model leaderboard and a plain-language benchmark guide), Free Tools (six tools — a model picker, an API cost calculator, a token counter, a VRAM calculator, a fine-tuning cost calculator, and head-to-head model comparisons), and Learn (the AI & ML glossary and the topic guides).

Curated reading paths through the blog's existing posts, found under Explore in the navigation or on the Topics page. There are five guides — RAG & Vector Search, AI Agents, LLMs in Production, How LLMs Work, and Machine Learning Fundamentals — each ordering its articles so every piece builds on the last, with pointers to the matching course and tools where relevant. Posts that belong to a guide also link back to it, so you can join a path from any article in it.

Yes — the AI & ML Glossary, under Explore, collects every term and symbol from the course glossaries into one searchable A-to-Z page. Type in the search box to filter terms as you type, and use the save button on any term to add it to your Flashcard Review deck.

An introduction, a quick look at the topics the blog covers, the most recently published posts, a way to ask the site's AI assistant a question, a preview of the AI model leaderboard, a set of free AI tools, a daily word puzzle teaser, a way to pick a course to start, and a list of older posts.

Yes. A floating sun/moon switch, always in the corner of the screen, lets you flip between light and dark mode, and your choice is remembered the next time you visit.

Yes, the site is fully responsive. On smaller screens, the navigation links collapse behind a hamburger icon in the top corner — tap it to open the menu, tap it again (or tap a link, or tap outside the menu) to close it.

Yes, once you've scrolled down a bit, a small "back to top" button appears in the corner of the screen. Click it to smoothly scroll back to the top of the page.

You'll land on a friendly "page not found" screen with quick links back to the homepage or the full list of posts.

Visit the Posts page for the complete, browsable archive. The homepage also shows recent posts and a preview of older ones, with links through to the full list.

Yes, the Posts page has a search box that matches against post titles, topics, and keywords as you type.

Yes, the Posts page offers topic filters such as RAG, LLM, Agents, Embeddings, MLOps, Fine-Tuning, Prompting, Regression, and Observability, each showing how many posts match.

It takes you straight to a randomly chosen post — a good option if you're not sure what to read next.

Yes, posts you've opened are marked as read, and posts you've explicitly marked complete are shown as done, so you can pick up where you left off.

Yes. Each post listing has a heart icon — click it to save that post to your reading list. To see everything you've saved, open the account menu in the navigation bar and choose "My Favourites."

A "See more" button loads additional posts a few at a time and shows you how many you've viewed out of the total. A "Show" selector above the list lets you choose how many posts appear per page.

Each post listing shows an estimated reading time (e.g. "5 min read") so you can pick something that fits the time you have.

Topics range from beginner-friendly machine learning fundamentals to more advanced subjects like language models, fine-tuning, and running AI systems in production. Each post is labeled with a difficulty level — Beginner, Intermediate, or Advanced — so you can choose content that matches your experience.

Yes — every post includes a "Listen to this post" player at the top. Press play to start, and the text currently being read is highlighted as it goes so you can follow along. While it's playing you can pause, jump to the previous or next section, or drag the progress bar to skip ahead. A settings icon lets you change the playback speed or pick a different voice.

Yes — near the top of every post you'll find an "A-" and "A+" control. Click "A-" to shrink the text or "A+" to enlarge it; there are five size steps in total, and whichever size you land on is remembered the next time you read a post.

Click the "Reading mode" button near the top of a post to switch to a simplified, distraction-free view that hides extra page elements so you can focus purely on the text. An "Exit reading mode" button appears while it's active — click it to return to the normal view.

Yes — for posts with several sections, a table of contents appears alongside the article (next to the text on wider screens), listing each heading as a clickable link. Click any entry to jump straight to that section, and the list highlights your current section automatically as you scroll. On narrower screens, it instead appears as a "Contents" dropdown you can tap open near the top of the post.

Yes, code examples include a one-click "Copy" button.

Yes, hovering over a section heading reveals an option to copy a link straight to that part of the article.

Those are glossary tooltips. Hovering, tapping, or focusing on a highlighted term shows a short, plain-language explanation without leaving the page.

Yes — click the "AI Summary" button near the top of a post, and a panel opens with a short, bullet-point overview of the article. It takes a few seconds to generate the first time; clicking the button again simply opens or closes the panel without regenerating it.

Yes, under "Found this useful?" near the end of each post you can pick from a few quick reactions — Helpful, Insightful, Great read, Practical, or Worth sharing. Clicking one highlights it to show it's selected; clicking it again un-selects it. Your picks are saved per post, so you'll see the same reactions highlighted if you revisit later.

Yes — near the top of each post, you'll find a small "Share" row with icons for X (Twitter), LinkedIn, WhatsApp, and Facebook. Click any of them to share directly to that platform, or click the copy-link icon to copy the post's URL to your clipboard.

At the end of a post, you'll often see suggested related articles chosen based on shared topics, so you can keep exploring similar content.

Yes, posts that involve formulas or statistical notation display them clearly within the article.

Yes, many posts end with a short "Test yourself" quiz. Click the quiz header to expand it and reveal the questions; clicking the header again collapses it.

Each question gives you a few multiple-choice options to pick from. After selecting an answer, click "Show answer" to instantly see whether you were right, with the correct option highlighted and a short explanation underneath. If you'd rather answer everything first, you can fill in all the questions and click "Submit" at the bottom to grade them all at once. Once a question is graded, you can click the same button (now labeled "Hide answer" / "Show answer") to toggle the explanation open or closed at any time.

A running score (e.g. "3/5") is displayed at the top of the quiz as you answer questions. Once you've answered every question, you'll also get an overall message reflecting how you did — anything from "Perfect score" to "Worth a re-read."

Yes, a "Reset" button clears all your answers so you can go through the quiz again as many times as you like.

Yes. As you scroll through a post, a progress bar at the top fills in to show how far you've read.

At the very bottom of a post, click the "Mark as complete" button. It changes to a checkmark and "Completed" once clicked. This counts toward your reading streak, and shows up as "Done" next to that post if you look it up later on the Posts page.

It's a simple counter that goes up when you read a post on consecutive days, encouraging regular reading. Once you've read on at least two days in a row, a streak badge (e.g. "🔥 3-day streak") appears around the site, with a small celebratory animation the first time you see it each day.

It offers full, structured courses built end to end as one curriculum, not a reading list of existing posts. Each course is organized into phases and modules, with original lessons, hands-on labs you run directly in your browser, and checkpoint quizzes.

Two courses. Machine Learning Fundamentals (Beginner) covers classical machine learning end to end: foundations, data wrangling and preprocessing, regression, evaluation, classification, unsupervised learning, model quality, and reinforcement learning, finishing with a capstone project. Deep Learning (Intermediate) picks up from there: neural network foundations, forward and backward propagation, optimization and regularization, CNNs, RNNs, and attention and transformers, plus practical skills for training and debugging networks — also finishing with a capstone project. More courses are planned.

Go to the Courses page and either pick a course from the "Where do I start?" picker at the top, or scroll down to browse the full curriculum. Each course lists its phases in order, with a "Start" button that takes you straight to the first lesson.

Each page shows which item you're on (e.g. "Lesson 3 of 5"), along with "Previous" and "Next" buttons at the bottom so you can move through a course in order. A course outline drawer is also available from the top of every page if you want to jump anywhere directly.

At the bottom of each lesson or lab, click "Mark as complete." The button updates to show it's done, and a checkmark appears next to that item everywhere it's listed, including the course outline and the phase listing on the Courses page.

Yes — course lessons end with a "Check your understanding" box. Click Start, explain the lesson's key idea in your own words, and click "Get feedback": an AI tutor replies in seconds with what you nailed, what's worth revisiting, and a short follow-up to take the idea further.

Yes. Once you've answered every question in a checkpoint quiz, it's automatically marked complete, the same way a lesson or lab is, and counts toward that phase's and the overall course's progress.

Each phase on the Courses page shows a progress bar and a count like "3/5 completed," updating as you finish lessons, labs, and quizzes. Phases are collapsible, and the page automatically opens whichever phase you haven't finished yet so you can pick up where you left off.

A banner appears on the Courses page letting you claim your skill badge for that course. You can personalize it with your name, and it stays on your Achievements page as a shareable credential.

A daily spaced-repetition flashcard deck, found under Flashcard Review in the account menu or from the My Progress page. It's built from the courses: every notation card and key term from the course glossaries, plus any checkpoint-quiz question you've answered wrong. Short daily sessions bring each card back right around the time you'd otherwise forget it.

Two places. Glossary cards cover every term and symbol from each course's glossary page, introduced gradually — at most ten new terms a day, so your first session isn't overwhelming. Missed-question cards are created automatically when you answer a checkpoint-quiz question wrong while signed in, so your deck naturally focuses on the things that actually tripped you up. You can also save any term from the AI & ML Glossary page as a flashcard with one click.

Each card climbs a ladder of review intervals. Answer "Got it" and the card moves up a step, returning after progressively longer gaps — one day, then three, seven, fourteen, and finally thirty. Answer "Again" and it drops back to the start and repeats within the same session until you get it right. Small dots on each card show how far it has climbed.

Tap the card (or press Space) to flip it over and reveal the answer, then grade yourself honestly: "Again" (or press 1) if you didn't know it, "Got it" (or press 2) if you did. On a touch screen, you can also swipe the revealed card — left for "Again," right for "Got it." If you mis-grade a card, an Undo button takes back your most recent answer. A progress bar tracks the session, and a summary at the end shows how many cards you got right on the first try.

If you're signed in, yes — your review schedule and missed questions are saved to your account and follow you across devices, the same way course progress does. Signed out, you can still review the glossary decks, but the schedule stays on that device only and missed quiz questions aren't collected.

A regularly updated comparison of popular AI language models, ranked by performance on general knowledge, coding, and math benchmarks.

It's a highlight of the current top 3 models, shown as cards with their overall score and a breakdown of their individual general knowledge, coding, and math scores — a quick summary before you scroll down to the full table.

By default, models are ordered by a combined score made up of three areas — general knowledge, coding ability, and math ability — with the highest combined score at the top. The top-3 podium at the top of the page always reflects this combined score, regardless of how you've sorted the table below it.

Yes. You can filter the table to show only open or only closed models, and you can click any column header (general knowledge, coding, math, or the combined score) to reorder the table by that score instead of the default ranking.

It refreshes weekly (every Monday) to keep the rankings current.

Yes, a "Retired" column flags models that have since been discontinued by their makers.

The leaderboard links to a companion guide, "The Benchmarks, Explained", which covers what MMLU measures, what the coding and math scores test, how the composite index is calculated, and why benchmark numbers should be read with some care.

Six, all under Explore in the navigation or at the Tools page: a model picker that recommends a model from four quick questions, an API cost calculator that prices your workload across 20+ models, a token counter that estimates tokens and checks context-window fit, a VRAM calculator that tells you whether a GPU can run an open model locally, a fine-tuning cost calculator that estimates what it costs to train and then serve a fine-tuned model from your dataset size and monthly usage, and head-to-head model comparisons pairing benchmark scores with pricing.

Completely free, no account or sign-up needed. They run entirely in your browser; the token counter in particular never uploads your text anywhere.

Benchmark scores come from the site's weekly-refreshed AI leaderboard, and prices are each provider's published API rates, updated on the same weekly schedule. Open-weights models show representative hosted prices, marked with a tilde, since rates vary by provider.

The model picker. Answer four questions about your task, budget, open-weights preference, and context needs, and it recommends three models with links to compare them in depth or price them for your workload.

A daily word puzzle, in the spirit of Wordle, themed around computer science and AI vocabulary. Guess the 5-letter term in 6 tries — everyone gets the same puzzle each day. It's reached from a small icon in the navigation bar.

Type a valid 5-letter word and hit Enter to submit it. After each guess, the tile colors tell you how close you were: a tile turns one color if that letter is correct and in the right spot, another color if it's in the word but the wrong spot, and stays plain if it's not in the word at all. You get 6 guesses to find the word, and you can't submit the same guess twice in one puzzle.

Yes, one new puzzle releases daily and it's the same for every player. A countdown to the next puzzle is shown in the Statistics panel.

Yes, tap "Hint" for a short definition of the word. You only get one hint per puzzle, so it's worth saving until you need it.

Yes. It tracks how many puzzles you've played, your win rate, your current and longest streak, and your guess distribution, all viewable from the "Stats" button. Signing in lets those stats follow you across devices; otherwise they're kept on your current device only.

Yes, the "Share" button in the Statistics panel lets you share your result without spoiling the word for others.

Yes, the Settings panel offers Hard Mode (any revealed hints must be reused in later guesses), a dark theme, a high-contrast color scheme for colorblind accessibility, an option to use only the on-screen keyboard (helpful for assistive devices) or your device's own keyboard instead, toggles for sound effects and background music, and works the same way on mobile as on desktop.

Ore is a friendly AI assistant available on every page of the site, shown as a small chat icon you can click to ask questions.

Ore can help with questions about the site's content, Peri's background and work, and how to get in touch. It's focused specifically on this site and won't answer unrelated questions.

Ore uses AI to understand your question and help you find relevant information on the site. Further detail about how it works internally isn't shared.

Ore can give more relevant help while you're on a specific post — for example, you can ask it to summarize the post you're reading.

Yes. Tap the microphone icon to speak a single question — Ore will transcribe it for you. If you'd rather have a hands-free back-and-forth conversation, press and hold the microphone icon briefly to switch on "voice mode": Ore will then automatically start listening again each time it finishes replying, until you hold the icon again to turn it off.

Yes, click the speaker icon on any of Ore's messages to have that reply read aloud; click it again to stop. If voice mode is switched on, Ore reads its replies aloud automatically.

Things like "What should I read first?", "Posts about RAG?", "How do I get in touch?", or "What does Peri work on?"

If something goes wrong, Ore will let you know and suggest reaching Peri directly by email or LinkedIn instead.

Yes — a small easter egg. Type the letters "o-r-e" anywhere on a page (outside of a text box), or click Ore's avatar in the chat header, and you'll get a fun animated celebration with a spoken greeting.

Yes. Ore is designed to stay focused on helping with the site, no matter how a conversation is phrased.

Yes, you can subscribe using the newsletter signup form with just your email address. It's an occasional email covering AI and ML topics, from fundamentals to the latest developments — no spam, and you can unsubscribe anytime. You'll see a confirmation once you're successfully subscribed.

Yes, the site publishes an RSS feed that any feed reader can subscribe to, so new posts show up there automatically.

Things like your theme preference, text-size preference, reading streak, and which posts you've read or completed are remembered on your own device so your experience stays consistent between visits. None of this is required just to browse and read the site.

Yes, basic website analytics help Peri understand which content is most useful to readers.

Reach out through the Contact page form, or email Peri directly. She reads every message.

Yes, all content on the site is free to read, with no subscription or paywall.