About Me
Hi, I’m Ashish Kulkarni! I’m a Computer Science graduate from PES University, and a Software Developer at Nasdaq. This site is a platform for me to showcase some of the work I’ve accumulated over the past few years: projects, articles, and more. Use the tabs above, and scroll around to see some of my work!
My Boring Story
I decided to pursue my undergraduate studies in Computer Science due to my affinity towards logic and mathematics. However, my first true experience with building software was only after I started at PES University.
In my first semester, I took part in a virtual hackathon. This hackathon was the first time I wrote code with any real purpose, unlike the assigments I had completed in the past. Within those 24 hours, I learnt about API calls, taught myself Python, and managed to snag a Top 5 position. You can check the project out here. The hackathon fueled my desire to build more complex systems, and pursue research.
The first experience I had with research was at a startup called StanceBeam. I interned at StanceBeam for two months in 2022, between my 4th and 5th semesters. StanceBeam is a sports technology company with an ambitious mission: to disrupt the cricket decision-review systems of today with a far cheaper alternative. As a computer vision intern, I worked on spatial positioning using stereovision and epipolar geometry. The work I did during these two productive months was demoed to BCCI in Dubai. This kind of exposure, early in my career, primed the way I approach problems today.
Around the same time, my interest in deep learning grew. I was especially interested in the way inputs can be preprocessed to affect scores, and how a neural net’s architecture can change it’s effectiveness in performing a task, in this case the detection of glaucoma using a retinal fundus image. A friend of mine, student of medicine, and I worked on a comparative study that looks at
- the difference in performance between CNN architectures, and
- the effect histogram equilization based preprocessing techniques have on the final scores.
You can read more about this project here, or read our preprint paper.
By the time I reached my 6th semester, genertive AI was all the rage. ChatGPT had become a household name, and new generative models were being released seemingly every single day. My friends and I were very interested in image and video generation models, and decided to pursue researching this as part of our capstone project. We had the privilege of presenting an approach to video generation as a poster at C-MInDS, IIT Bombay. We messed around with diffusion-based text-to-image models, and came across a parameter classifier-free guidance scale, which we found had a significant impact on the generated output. Read more about our research here.
After my 6th semester, I joined Nasdaq as a summer intern. For my internship project, I was tasked with creating an NLP-based minimum viable product, and after 2 months, pitched it to the product team. The product was greenlit, and my internship duration was extended to allow me to continue working on it.
Over the next 11 months of my internship, I took the product from MVP, to prototype phase, to its first release as the primary Python developer. This project was the perfect fit for me, as I was given an R&D role, with tons of ownership over the product. Apart from the majority of backend development, I played an important role in database schema designing, creating a cloud-native architecture and provisioning resources using Terraform on AWS, and evaluating NLP and LLM models for very specific use-cases. Today, I work as a full-time software developer on the same project. My day-to-day varies between developing backend process, GenAI integration, and most recently, implementing cloud-based malware protection.
During my 8th semester, along with my internship, I served as an undergraduate teaching assistant for a 6th-semester course for 4 months, you can read more about that here. In parallel, I mentored at a 6-month long machine learning bootcamp at Nasdaq, Bengaluru.
