Hello, I'm Christian Miller

Into the Metaverse

As I settled into my second day at Aeye Inc, a computer vision startup, I noticed a sticky note with 'Pointcloud in VR' scribbled on it. Following discussions with my mentor to set project goals, I embarked on the development of a VR application, aiming to enable users to visualize and interact with dynamic LiDAR pointcloud data.

After a week of thorough research, I found no evidence of anyone attempting this with dynamic data. Undeterred, I drew inspiration from techniques outlined in various research papers. Leveraging the capabilities of the Unity Game Engine and an HTC Vive 3, I set out to accomplish the task.

A couple of weeks later, a working prototype emerged. Despite its initial capabilities to load and navigate through pointclouds, I understood that optimization was pivotal for an enhanced user experience. Recognizing the potential, I strategically leveraged the multithreading and concurrency capabilities of the GPU in combination with Unity. This not only optimized the application's performance but also laid the groundwork for real-time data viewing from a sensor—an achievement that brought the project one step closer to a seamless and dynamic user interaction. After presenting to the company, I had a line of people from all backgrounds coming by to try it on for themselves. Rendering over 1000 frames comprised of 2GB csvs each in seconds, the application was a success.

In the end, what started as a scribble on a sticky note unfolded into a transformative journey. The optimized VR application not only showcased the power of computer vision but also paved the way for real-time data visualization. This experience at Aeye not only sharpened my technical skills but reaffirmed my belief in the potential of pushing boundaries within the realms of technology.

• See it on github!

Predicting NBA Games

As my data structures class concluded, our final assignment beckoned: implement a data structure with practical use. Enticed by the allure of riches and terminonology such as machine learning, my team embarked on what turned out to be a challenging journey.

The following weeks blurred into a whirlwind of late nights, tutorials, and manuals, all within the confines of the windowless basement of the engineering building. Despite the frustration, we persisted and pieced together the necessary components. We collected and cleaned data, trained a machine learning model for classification, and implemented a minimum heap to fulfill the class requirements, showcasing the 'closest' games.

The culmination was a functioning NBA game predictor boasting an impressive 74% accuracy. Built utilizing PyTorch and Python, the model was trained on a dataset of over 7000 games from the past several seasons. Feature engineering played a crucial role, extracting insights from player statistics, team performance, and historical match data.

The prediction algorithm considered various factors such as team momentum, statistics, and home-court advantage. Continuous refinement of the model involved tweaking hyperparameters and which statistics best represented a team's performance, enhancing its accuracy over time.

Our journey, though initially challenging, proved rewarding. The project not only demonstrated our proficiency in implementing sophisticated data structures but also highlighted the potential of combining machine learning and classical algorithms in predicting real-world scenarios. As we unveiled our NBA game predictor to our professor, the windowless basement transformed into a space of triumph, a testament to the determination and collaborative spirit that fueled our success.

• See it on github!

BetND

In the fall of 2022, I embarked on a journey to pioneer BetND, an on-campus sports betting marketplace that would soon become a hub for over 60 daily active users. This project not only showcased my technical skills but also highlighted my ability to lead and collaborate with a team of 4 peers.

One of the project's key innovations was the implementation of the ability to transact current bets with other users. For example, if a user placed a bet on a team to win, then halfway through the game the team was up by 20 points, the user could sell the bet to another user for a profit.

To support the robust functionalities of BetND, we engineered a comprehensive database. Comprising over 8000 entries distributed across 7 tables, the database was designed for organized and efficient data manipulation and access using SQL. This architecture laid the foundation for a reliable and scalable system that could handle the intricacies of sports betting data.

The culmination of our efforts resulted in the creation of a modern and user-intuitive multi-paged website. Developed using a tech stack that included PHP, JavaScript, and Python, the website presented users with all the relevant information they needed for informed and enjoyable betting experiences.

BetND stands not only as a testament to my technical prowess but also as a demonstration of effective collaboration, innovation, and entrepreneurial spirit. This project has become a cornerstone in my portfolio, showcasing my ability to conceptualize, develop, and successfully launch a complex, data-driven platform.

• See it on github!