Getting back to Machine Learning in 2025

Okay. So AI is definitely super hyped and I can't help but get one of the biggest FOMO of my life. LLMs, MCP, RAG, transformer this attention that. There is just so much chatter about AI these days on the internet, it feels like everyone is an expert in the field. Back when I was into machine learning during my university days, the coolest things happening with AI were mostly research oriented, like fancy new model architectures, and training techniques. But post GPT-3, there is one aspect of AI that has widely been opened up: engineering. Before GPT, it was quite hard to imagine how to engineer day-to-day products or real-life use cases of AI models. But after LLMs happened, and big tech companies came up with better models, intelligence is readily available to all the engineers simply via an API!!

As an engineer, I am quite excited to start playing around with this new piece of technology. I do have a machine learning educational background, but after I went down the crypto smart contract engineer rabbit hole, I have mostly lost touch. I am looking to spend some time studying something new relevant to the field every week, and post about it on my blog. I don't want to start directly with an LLM API and build another chat application. I am genuinely interested in going one level deeper and start from understanding the inner workings of LLMs and how we really got here.

This blog is about leaving a trail of my journey towards the future and keep documenting my thoughts on the subject.