Blog

What this blog is

I am a senior engineer working on a multi-LLM platform in production, and I also run @9_face_toon, a Hindi animation page with around 40K followers each on Instagram and YouTube. Almost everything here comes out of one of those two jobs. When I write about React 19 transitions, JWT refresh flows, or Vercel middleware, it is because I hit the problem at work and had to solve it. When I write about video or voice tools, it is because they are in the pipeline that ships my own content every week.

That also defines what you will not find here. I do not publish roundups of tools I have not used, and I do not rewrite launch announcements into articles. If a product is interesting but I have not run it on real work, I say so rather than inventing an opinion — you will see that stated explicitly inside several posts where I have only tested something briefly.

The four things I write about

• Next.js and React in production — image and LCP tuning, App Router sitemaps, React 19 transition hooks, and edge middleware versus static redirects, each written from a change I actually shipped.
• Backend and agent architecture — JWT authentication done properly with refresh tokens and HttpOnly cookies, and a comparison of two agent stacks (AWS Bedrock with A2A/MCP versus LangGraph) I built for different clients.
• AI tools, tested on real work — coding editors, local models, research tools, and image models, judged on whether they survive a normal working week rather than a demo.
• Creator and freelance economics in India — how the animation channel grew, which tools are in the actual production pipeline, and what realistic rates look like when you are starting out here.

How each article gets made

A post starts when something takes me longer than it should have. I write down what broke, what I tried, and what finally worked, then check my claims against official documentation, release notes, and pricing pages before publishing. Where a post relies on numbers I measured myself, I say how I measured them and where that measurement is weak — a single developer on one machine is not a controlled benchmark, and articles here label it that way rather than implying more rigour than exists.

I use AI assistance the same way most developers do: for outlining, tidying grammar, and pressure-testing an argument. The examples, the mistakes, the code, and the judgments are mine, and nothing is published as a raw generated draft. Where a claim can be verified, most articles carry a sources-and-verification block at the end listing the primary pages I checked and the date I checked them.

Full testing, sourcing, AI-assistance, disclosure, and corrections standards are in the editorial policy. If you find something wrong in a post, tell me — corrections get made in place with a note, not quietly deleted.