Basically, Github Copilot that can run locally on a laptop.

1. Built a custom Gemma 2 270M variant.
2. Utilizing the knowledge distillation, trained the base model first on english dataset (FineWebEdu) using Gemma 2 2B as teacher using the classic next word prediction technique.
3. In the second training phase, used CodeGemma 2B as teacher to train the pretrained 270M base model on codecontest dataset using the Fill in the Middle technique.

Technologies Used:

Teras is a unified deep learning library for Tabular Data that aims to be your one stop for everything related to deep learing with tabular data.

Checkout the official repo, https://github.com/KhawajaAbaid/teras/

1. It provides state of the art layers, models and arhitectures for all purposes, be it classification, regression or even data generation and imputation using state of the art deep learning architectures.
2. While these state of the art architectures can be quite sophisticated, Teras, thanks to the incredible design of Keras, abstracts away all the complications and sophistication and makes it easy as ever to access those models and put them to use.

Technologies Used:

Try this out in a browser, https://khawajaabaid.github.io/prewordict

1. Processed 180,000 medium articles and clustered the similar words
2. Generated word clouds from these clusters. Given a word cloud, user has to predict one word that best belongs in the word cloud.

Technologies Used:

Consists of three main modules:

1. FPS Predictor: Predicts the expected amount of FPS for a user-specified video game on a user specified PC configuration.
2. Games Recommender: Recommends video games based on user specified PC configuration. Uses a simple yet working ranking algorithm. (currently only CPU and GPU)
3. PC Predictor: Recommends PC components based on user specified video games. Uses pretty much the same ranking algorithm but in reverse. (currently only CPU and GPU)

Technologies Used:

1. Worked on Computer Vision to train Deep Learning models for Person Detection as well as Detection of Certain (similarly attired) Groups among crowd.
2. Developed custom algorithms for Chaos Detection and Crowd Detection

Technologies Used:

1. Scraped games info and their FPS data from the internet (Steam and other sites)
2. Used Random Forest algorithm to build the FPS predictor

Technologies Used:

Check this out here, https://twitter.com/TweetsCloudBot

1. Uses Twitter API to get 100 most recent user tweets
2. Generate a user-customizable tweets cloud (word cloud of most frequent words) in tweets, filtering the stop words.

Technologies Used: