Langformers

08/05/2026

I learnt something interesting today about user behavior and it challenged one of my long held beliefs.

I have always loved minimal blog designs. Clean layout. Single column content. No sidebars. No distractions. Just good writing. That philosophy shaped the entire "Langformers Blog". People were reading the articles for a long time, traffic kept growing steadily through Google, Bing, LinkedIn, X, and even AI chatbots. Everything looked fine on the surface. But there was one thing I completely overlooked. Readers were consuming the article... and leaving (i.e., unique visitors ≈ pageviews). Not because the content was bad; some of the messages I receive genuinely make my day. Looking at it now, readers probably just needed a small nudge to discover more content on the site.

So I tried a tiny experiment.

After someone finishes an article and reaches the comment section, a small recommendation popup appears (attached image) with a few suggested articles. Nothing aggressive. No interruptions while reading. Close it once and it stays gone.

That single change DOUBLED the pageviews. 😎

Funny enough, the lesson was not about popups or metrics. Minimalism is great, but if you remove every pathway, people cannot explore.

27/04/2026

In April 2025, I released a Python library "Langformers". At the time, it was mostly an idea I wanted to explore around working with LLMs and MLMs in a simpler way. Since then, it has quietly grown to 8.5k+ pip installs, which is honestly more than I expected when I first pushed that 0.1.0 version.

Over the past year, I have had a lot of ideas for where to take it next. Many of them exist as rough code sitting locally, some half-finished, most without tests. But like many side ventures, it has had to compete with full-time work and other responsibilities.

One thing that did grow consistently is the "Langformers Blog"; which is now getting 3k+ unique readers a month, and that has been really motivating. Writing about LLMs in a simple and understandable way has helped me clarify my own thinking as much as it has helped others.

I see Langformers as still being at the beginning. There is a lot I want to build, clean up, and properly release. I hope this next year is about being more consistent, releasing smaller pieces, and making the library even more useful.

If you have used Langformers, read the blog, or even just shared feedback, thank you. It really does make a difference. 🙏

24/04/2026

It was about time I wrote this article. Everyone seems to be talking about LLM agents.

I’ve personally seen a huge number of LLM agents being built for different domains and data. I think knowing at least the basics, which can later be generalized to more complex scenarios, is worth it.

If you’ve ever thought one of the following, this guide is definitely for you:
-> I understand Python, but LLM agents still feel confusing.
-> I hear terms like agent, tools, graph, state, prompt, retry, planner, executor, but I don’t really get it yet.
-> I want one example that is simple enough to understand, but real enough to use in actual projects.

Article:

It was about time I wrote this article. Everyone seems to be talking about LLM agents. I’ve personally seen a huge number of LLM agents being built for different domains and data. I think knowing at least the basics, which can later be generalized to more complex scenarios, is

11/04/2026

Well, this is encouraging. In the last 30 days, over 3k visitors (bots excluded) stopped by the "Langformers blog". These are the pages that got read the most.

What stood out to me is that most of the interest is still around the fundamentals: how LLMs work, embeddings, gradients, chunking, etc. I honestly didn’t expect that. I thought the more applied stuff would lead.

I haven't been actively pushing content lately, so seeing this kind of organic interest is really nice. If you've read, shared, or found something useful there, thank you. It genuinely means a lot.

🔗 Langformers Blog: https://blog.langformers.com/

25/02/2026

LLMs are no longer just a niche topic for CS or Math guys; they’re everywhere.

I looked at Google Search Console data from the last 28 days to understand what people are actually searching for around LLMs. Here are some Langformers blogs receiving the most impressions (impressions = appearances in search results):

19.4k - BPE Tokenizer: Training and Tokenization Explained
11.9k - How LLMs Work: A Beginner's Guide to Decoder-Only Transformers
8.8k - A Step-by-Step Forward Pass and Backpropagation Example
5.2k - Ollama and Local LLMs: Step-by-Step Guide
3.8k - Fine-Tuning LLMs on Custom Data with LoRA
3.3k - Chunking Strategies for LLMs (Fixed-size, Semantic, Recursive)

I’ve been seeing a similar upward trend over the past few months.

What this suggests:
- Strong interest in fundamentals (tokenization, transformers, backprop)
- "How things work" content is still rocking

Feels like we’re moving from "hype" to "understanding".

05/02/2026

🚀 Just published an extremely beginner-friendly guide explaining the core idea of fine-tuning an LLM in the easiest way possible.

📖 Article: https://blog.langformers.com/train-llm-custom-data/

LoRA has been around for a while, and I’ve been working on bringing LLM fine-tuning on custom data into the Langformers library. Once testing is complete, the feature will be released very soon. :)

Large language models (LLMs) can feel a little magical at first. You ask a question, hit enter, and almost instantly an answer appears. Sometimes it's impressively accurate. Other times, not so much. And occasionally, the model gives a response that sounds completely convincing… even though it's e...