Asif's Profile

Personal Website (this one!!) developed using FastAPI (with Jinja2 templates) and SQLAlchemy (with SQLite database), and self-hosted using Docker, with proxy server using Nginx. Current architecture of the website has provision to support multiple users, with each user having their own profile. For simplicity, it currently supports only my profile, and adding a more streamlined database update pipeline is on the way. Also built an automated pipeline with GitHub Actions to build the Docker image and push it to Docker Hub to be deployed on every push to the main branch with automated unit tests written with PyTest. The website can create timeline dynamically from education and experiences stored in the database for the user, and alters theme based on user's preference.

Reproduced the research of Wang, Ji (2021) et al. Converted EEG signals to text tokens using pre-trained BART model and used zero-shot classification algorithm to verify the sentiment of the EEG signals. The resulting algorithm performed on par with the original research in most metrics and outperformed it in some.

Constructed a novel architecture for object tracking using FairMOT and GAN and demonstrated that using generator in a separate layer of the architecture makes the tracking performance much worse because the discriminator learns from the layer itself if the data is from fake distribution or original.

Fine-tuned GPT 3.5-turbo 1106 model with prompt engineering from historic game events coupled with live commentary data to generate realistic commentaries and game summaries for new games. Also leveraged text-to-speech and translation APIs to emulate voices of famous commentators like Peter Drury for a more realistic feeling in the generated commentary. The translated commentary, along with closed captions are aimed to be used to make experiencing the game more accessible.

Used Tensorflow to recognize various objects in images and videos from a mobile app with on-device AI model, and optimized the algorithm by implementing GPU based execution with CUDA in C++. Developed during my bachelors (2018), the accuracy of the method turned out to be over 70% for recognizing everyday objects and detecting faces in video streams.

Tried hands-on learning approach with Unity3D game development engine by building a clone of popular side-scrolling game Flappy Bird, with personalized object models built with Blender and custom physics inside the engine. The game, even though developed well, did not have a very aesthetic appeal due to the unavailability of professional assets and sounds, and consequently was not released in any game store. But this experience gave me a headway into game development along with experience with game engines, Blender and C# language during my bachelors.

Built scalable ETL pipeline using PySpark to preprocess, filter gigabytes of genomic data and optimized query performance with partition pruning and caching to reduce query time by 50%. Also implemented MLlib-based clustering (PCA + K-Means) to analyze mutation frequency and gene correlation.

Developed a CLI tool in Python to align DNA sequences using dynamic programming algorithms like Smith-Waterman (local sequences) and Needleman-Wunsch (global sequences). Used libraries like Numpy for efficient computation, Pandas for data manipulation along with various visualization options (CLI-based color-coded alignments, ASCII heatmaps and Matplotlib). The input system was developed to take in arguments from CLI as well as from JSON, YAML, CSV or Text files. The tool was developed with over 90% test coverage with PyTest with a modular architecture in mind, and was used to align sequences for a bioinformatics project during my bachelors.

Developed an automated data ingestion pipeline using third-party APIs (Twitter, Reddit) in Python to fetch posts specifically related to AI and Technology, managing data retrieval with GitHub Actions. Applied sentiment analysis with cloud-based big data services, leveraging PySpark for parallel processing, secure storage via Azure Storage, and scalable analytics and processing using Databricks Job Scheduler. Built a user-friendly interactive dashboard with Streamlit, deployed via GitHub Actions as a real-time web app in Azure. Ensured best practices for version control, security, and scalability.

Learned how to use NumPy, Pandas, Seaborn, Matplotlib, Plotly, Scikit-Learn, Machine Learning, Tensorflow, etc with respect to ML Projects.

Certificate earned on 2021-05-06

Learned how to build Shallow and Deep Neural Networks using Scikit-learn and PyTorch.

Certificate earned on 2020-01-16

Learned how to improve Deep Neural Networks using regularization, dropout, batch normalization, etc. Dove deep into the black box of deep learning to understand how and why it works so well.

Certificate earned on 2020-07-15

Hands-on approach towards learning to use PyTorch for practically developing and building deep learning models. This complements my learning in the classroom for classes like Computer Vision, Natural Language Processing and Machine Learning, where I often designed the models, but wanted to get a more hands-on flavor towards working in the field.

The certification validates the holder's knowledge of the core concepts of the Apache Spark and the ability to apply them to solve real-world problems. Ongoing learning to appear for the certification exam in upcoming months.