Hritij Rana

I'm a '24  Mathematics and Computing undergrad. from Netaji Subhas University of Technology. I am an ex-competitive coder and prompt engineer at softage.ai. I have a bittersweet relation with LLMs where sometimes it's fun to make it reason and other times its not. I was part of the team responsible for creating agentic data for a computer-use agent for general agents. These days I am on a lookout to join a close knit AI startup.

My areas of interest include on-device AI. I think it won't be long when LMs could run locally on our phones and PCs. We've already been seeing great effort by hardware giants in this area. Drop an email, and let's chat — even if it's just a silly idea!

Links: Linkedin • Twitter • GitHub • LeetCode • CodeChef

Hritij Rana

Projects  [O.S] • Blog • Awards

AI Judge

What would happen if you let AI reason for the hackathon winner?

The Story : An AI Judge that compares every other submission, taking submission video and description into consideration to reason and announce winner based on the judging criteria and winner category.It's a heads-on 1v1 battle among all the submissions ⚔️
Don't trouble yourself with the thoughts of who's gonna be the winner. sit back and let the 'AI judge' decide !

NameR : Gemini Nano powered PDF downloader

A file downloader that suggests file name on the go, so you never have to worry about absurd default filenames

The Story : The idea to built such a thing didn't come all of a sudden. There were countless times when I wasted hours searching for a specific PDF—simply because I hadn’t renamed it to something meaningful when I downloaded it. I’d end up scrolling through a chaotic list of files, unsure of their names. This frustration became all too common. The trigger point came when I was at a medical clinic, scrolling through my phone, searching for past medical records with absurd filenames that I never bothered renaming. That moment made me realize how a simple oversight can lead to significant inconvenience.

Privacy was a major concern as we don't want any sensitive records go for analysis. The project came to life when Google announced its Google's Chrome Built-in AI Challenge. This allowed us to leverage its on-device AI i.e Gemini Nano. We've utilized Gemini Nano to analyse the file locally and suggest a filename during download. The winners are due on January 13, 2025.

Named Entity Recognition for Automotive Vehicles

Fine-tuned SpaCy’s English model to perform Named entity recognition on tags like components, failure issue, vehicle model and corrective action.

The Story : This was part of an assignment where we needed to guess tags related to vehicle entites from an unstructured text log.

Submission to Mistral-AI Hackathon

One of the steps involved in cleaning a video is trimming unnecessary portion from it. For instance, expert labelled videos like 'Minecraft gameplay' don't require video segement of user setting up the game. This is a manual task. I've tried to automate this leveraing LLM and video transcript.

The Story : Being in the operations-AI industry for a while I realized that video cleaning is a tiresome task and one of the steps involved in video cleaning is to timestamp relevant portion of the video which is then used to train an AI Agent. So why not leverage LLM and video transcript for this task!