Reflections on 2025
As the year comes to an end, with the new year just an hour away, here is my very first blog. Let’s kick this off by going over my reflections on 2025.
Why do I want to start writing blogs all of a sudden? Well, two main reasons. In this era where everything is AI-generated and we run to LLMs for every small thing, I want to put my thoughts into words clearly—writing on my own and improving my clarity of thought, with AI help restricted only to grammar. The other reason is that I want a platform to share the cool things I am working on.
This year has been an absolute roller coaster.
The year started off with the beginning of our 6th semester, one of the most bittersweet semesters of college, got really close to a few friends and grew really distant from quite a lot of them.
I started off the year building Manimator, a paper/concept-to–3Blue1Brown-style AI animation generator. This started as a hackathon project, which we improved upon within two days of launching it. It ended up trending and was featured on Hugging Face Spaces of the Week, with thousands of people trying it out. Around the same time, our work was also accepted to an ICML workshop.
We then presented and concluded our Bosch University Connect Program project, which was an Agentic RAG-based vehicular troubleshooting assistance tool to help new technicians diagnose automobile issues faster and more accurately.
I started doing AI research at CognitiveLab on multimodal, multilingual document vision tasks as part of the Nayana project, for which we received a six-figure grant from Meta. The Nayana research papers we worked on were accepted and presented at NAACL, CVPR, and ICCV workshops.
Somewhere around this point, I realized that I really enjoy AI research and that research is probably the path ahead for me. This was mainly because, in research, there’s something new that comes out every day, you have to keep reading papers, keep yourself up to date, and you get to work on very recent, cutting-edge stuff.
Right around that time, I received a summer internship offer at IBM Research in the Quantum team.
I worked as a research intern at IBM Research for three months in the AI for Quantum team, where I got to contribute to the Quantum Support team and the Qiskit Code Assistant by working on agentic applications. I had a wonderful mentor, and this was honestly the best three months I could have asked for. I got to work at the exact intersection of two things I really like and want to work in: AI and quantum computing. I also worked on a patent during this internship.
Later in the year, Parichaya, my IIIT Bangalore research paper which I had co-authored, was published at CODS-COMAD. Our work on benchmarking QML models was accepted and presented at the AI-ML Systems Conference, Quantum Symposium.
I then became really interested in the area of using AI for the preservation and understanding of cultural heritage. Along with two friends, I wrote two papers in this space, one on fine-tuning LLMs for Sanskrit sandhi splitting, and another on fine-tuning VLMs for Bharatanatyam mudra detection. These were accepted and presented at EMNLP and ICCV workshops.
I also organized a Qiskit Fall Fest extension event in collaboration with IBM Quantum.
Our capstone project paper on distributed heterogeneous LLM inferencing, HyperCluster, was accepted to an AAAI workshop.
Towards the end of the year, I received a six-month internship offer from Microsoft Research to work in the RL for LLMs team, where I’ll be working on building deep research agents using rubric rewards.
These are the ups and the good things that happened, but there were obviously a lot of downs as well, which I won’t be talking about here. Bad things will always exist in life; it’s what we choose to focus on that matters.
This year taught me a lot of necessary things the hard way. Keep doing good, and good will happen to you.
At this point last year, I had absolutely no idea that this is how and where I would be right now. Let’s see what happens and where 2026 takes us.
Happy New Year!