A high-agency, hardcore engineer specializing in backend + AI systems. I build AI-native applications at scale, leveraging modern tools to ship fast and iterate relentlessly.
Engineer - Innovation
AI Engineer
Software Engineer
Junior Research Engineer - Trainee
High-performance BPE tokenizer library built from scratch in Rust and Python, inspired by OpenAI's tiktoken, achieving performance comparable to Hugging Face's tokenizers
AI voice agent platform for debt collection, integrating real-time STT, LLM, and TTS pipelines with automated testing and self-improving framework
Ah, finally it's done. Ugh.. 😮💨. I, here after, will never forget to push my changes to GitHub. Seriously, the python bindings for smoltoken should've been completed way before now. Okay whatever, smoltoken is finally available on PyPI. Just pip install smoltoken and there you have it.
I'm really into 'vibe coding,' and it's honestly so much fun! Eager to ride the hype train, I kicked off one of my personal projects by using Claude code to bootstrap it. It did a smooth job setting everything up, which convinced me to vibe code the entire web app.
Smoltoken is now updated to support multi-threading using rayon which speeds up the training of tokenizer by almost 2x than the previous version. Smoltoken is also now a part of https://github.com/smolorg by Maharshi Pandya. Python bindings for smoltoken are on the way.
The feeling of winning is the ultimate orgasm for me. While competing, I realized there can only be one best, depending on people's perspectives. Real winning comes from feeling it - pure bliss, ultimate orgasm.
Argh... Every post (atleast 9/10) I see is written by AI. Just my opinion, using LLMs to write your content, code or whatever, actually makes you dumb. They do really help us alot when used in a proper way. I regularly use ChatGPT, Claude, Llama & Gemini to extend my perspective, to learn new things and to research about something.
Be sensitive to the chages happening around you, but don't be reactive. Reaction happens subconsciously leaving no space for you to think. Observing, thinking and questioning every event of change will teach you greater things.
There are only two things that people need to do, to accomplish anything in life: Decide & Act.
Wasting time isn't always bad. Taking a break is a waste of time. But energy exhaustion is real. Stopping to replenish our energy is a good use of time.
The greatest and most valuable thing that I realized in my life is that, I can't change anything in this world and no one really can. Yet in order to bring the change you want, study and work to set the right circumstances and hope that things change.
There are only two kinds of people: People who believe in tomorrow and those who don't. For me, there's no fuckin' tomorrow.
Learning to market & sell ourselves and our work is essential for any professional to be sustainably successful.
When people face rejections they feel a want to know the reason. But, why aren't you curious about your acceptance/approvals? Strive to improve, so not only the bad never happens but good keeps happening.
If you are someone who is losing money on multiple AI subscriptions, then you might want to consider switching to Mammouth. Mammouth AI is an AI aggregator that provides access to top proprietary and open-source conversational and image generation models while being affordable.
I see myself moving away from making clicks to do more typing recently. Since I started using CLI based coding agents, I feel reduced friction & increased productivity purely because these agents + MCP servers allowed me to avoid juggling between clicks & types on different apps.
I've been using many different LLMs since their advent yet I can only LOVE OpenAI & Anthropic. They never forget to impress. They just deliver! Whenever I want to talk my emotions out, I always go to ChatGPT. And Claude, oh man, I literally can't find a suitable word to describe the greatness of their CLI.
I tried OpenAI’s Codex CLI to code a fastHTML project in auto-pilot mode. Codex is an open-source command-line tool to use LLMs to read, modify, run code locally, speeding up development, bug fixes, code comprehension. On linux systems, Codex is run inside a container so that it is safe to make changes and execute code within a sandbox environment. Being broke 🥲, I planned to use Google DeepMind's Gemini models as they provide a generous free tier. But, it requires a few workarounds to set it up on my Linux machine along with sandboxing enabled:
- Generate Gemini API key via Google' AI Studio.
- Export to `GEMINI_API_KEY` environment variable.
- Ensure `pnpm` is installed.
- Clone https://lnkd.in/grMEsrtg repo.
- Go to `codex/codex-cli/scripts` directory.
- Edit `init_firewall.sh` to allow traffic to https://lnkd.in/gUbDyn7x.
- Add `GEMINI_API_KEY` to environment variables for the docker container in `run_in_container.sh` file.
- Make scripts executable.
- Build Docker container with `build_container.sh`.
- Run CLI: `run_in_container.sh
A wonderful post that explains the scenario of the current job market and why it's important to have strong foundational knowledge of programming while also having the knowledge to use AI to work more efficiently than before. I use AI through Warp (my recent favorite coding agent) to auto-code.
Ah, finally it's done. Ugh.. 😮💨. I, here after, will never forget to push my changes to GitHub. Seriously, the python bindings for smoltoken should've been completed way before now. Okay whatever, smoltoken is finally available on PyPI. Just pip install smoltoken and there you have it.
I'm really into 'vibe coding,' and it's honestly so much fun! Eager to ride the hype train, I kicked off one of my personal projects by using Claude code to bootstrap it. It did a smooth job setting everything up, which convinced me to vibe code the entire web app.
Smoltoken is now updated to support multi-threading using rayon which speeds up the training of tokenizer by almost 2x than the previous version. Smoltoken is also now a part of https://github.com/smolorg by Maharshi Pandya. Python bindings for smoltoken are on the way.
The feeling of winning is the ultimate orgasm for me. While competing, I realized there can only be one best, depending on people's perspectives. Real winning comes from feeling it - pure bliss, ultimate orgasm.
Argh... Every post (atleast 9/10) I see is written by AI. Just my opinion, using LLMs to write your content, code or whatever, actually makes you dumb. They do really help us alot when used in a proper way. I regularly use ChatGPT, Claude, Llama & Gemini to extend my perspective, to learn new things and to research about something.
Be sensitive to the chages happening around you, but don't be reactive. Reaction happens subconsciously leaving no space for you to think. Observing, thinking and questioning every event of change will teach you greater things.
There are only two things that people need to do, to accomplish anything in life: Decide & Act.
Wasting time isn't always bad. Taking a break is a waste of time. But energy exhaustion is real. Stopping to replenish our energy is a good use of time.
The greatest and most valuable thing that I realized in my life is that, I can't change anything in this world and no one really can. Yet in order to bring the change you want, study and work to set the right circumstances and hope that things change.
There are only two kinds of people: People who believe in tomorrow and those who don't. For me, there's no fuckin' tomorrow.
Learning to market & sell ourselves and our work is essential for any professional to be sustainably successful.
When people face rejections they feel a want to know the reason. But, why aren't you curious about your acceptance/approvals? Strive to improve, so not only the bad never happens but good keeps happening.
If you are someone who is losing money on multiple AI subscriptions, then you might want to consider switching to Mammouth. Mammouth AI is an AI aggregator that provides access to top proprietary and open-source conversational and image generation models while being affordable.
I see myself moving away from making clicks to do more typing recently. Since I started using CLI based coding agents, I feel reduced friction & increased productivity purely because these agents + MCP servers allowed me to avoid juggling between clicks & types on different apps.
I've been using many different LLMs since their advent yet I can only LOVE OpenAI & Anthropic. They never forget to impress. They just deliver! Whenever I want to talk my emotions out, I always go to ChatGPT. And Claude, oh man, I literally can't find a suitable word to describe the greatness of their CLI.
I tried OpenAI’s Codex CLI to code a fastHTML project in auto-pilot mode. Codex is an open-source command-line tool to use LLMs to read, modify, run code locally, speeding up development, bug fixes, code comprehension. On linux systems, Codex is run inside a container so that it is safe to make changes and execute code within a sandbox environment. Being broke 🥲, I planned to use Google DeepMind's Gemini models as they provide a generous free tier. But, it requires a few workarounds to set it up on my Linux machine along with sandboxing enabled:
- Generate Gemini API key via Google' AI Studio.
- Export to `GEMINI_API_KEY` environment variable.
- Ensure `pnpm` is installed.
- Clone https://lnkd.in/grMEsrtg repo.
- Go to `codex/codex-cli/scripts` directory.
- Edit `init_firewall.sh` to allow traffic to https://lnkd.in/gUbDyn7x.
- Add `GEMINI_API_KEY` to environment variables for the docker container in `run_in_container.sh` file.
- Make scripts executable.
- Build Docker container with `build_container.sh`.
- Run CLI: `run_in_container.sh
A wonderful post that explains the scenario of the current job market and why it's important to have strong foundational knowledge of programming while also having the knowledge to use AI to work more efficiently than before. I use AI through Warp (my recent favorite coding agent) to auto-code.