Daniel Stabile

January 2022 - May 2022

For our final project we worked with a WidowX robot from Interbotix and trained it to flip pancakes! Our setup included 3 realsense cameras, AR tags mounted on pancakes, a custom 3D printed frying pan, and a cage surrounding the robot. By collecting rosbags of joint states and velocities during demo flips we were able to train a model that took in the joint/velocity states of the robot and the pose of the pancake (poses were obtained with AR tags) and outputted velocity commands for each joint. Here's a video of our project: Pancake Flipping Robot

August 2021 - December 2021

This was another project done with MIT Lincoln Laboratory and other Cornell students. The goal was to train a model that would take in SAR satellite data, identify vessels, and categorize them as IUU or not. IUU fishing is a problem often associated with drug/human trafficking and other illegal activities not to mention it can wreck ecosystems and often is done in waters of developing nations who don't have the resources to combat it. This project was for the xView3 challenge and while our team did not perform well, it was a huge learning lesson for me. If nothing else it, taught me a lot about working in cross-institutional teams and I like to think that the experience will be very helpful in my future career.

June 2021 - August 2021

This summer I interned at MIT Lincoln Laboratories and worked on a project focused detecting anomalous flight activities using machine learning. I can't comment much on this project for security reasons, but my core contribution was developing a pipeline generate synthetic data which could be used for training. The problem with detecting anomalous flight activities is that they're anomalous... and hence rare. Using synthetic data helps create example data for models to train on!

August 2020 - December 2020

In the fall of 2020, I took ENGRC 3025 Creating and Communicating Your Digital Professionalism taught by Dr. Allison Hutchison and Dr. Rick Evans. For the final project, I created a professional website using HTML, CSS, and the bootstrap framework. The website is built upon a template but was significantly altered to fit my rhetorical situation. The most notable changes from the template are the projects and travel sections. I wanted to keep the site to one page so for the projects section I created modals with more information for each project listed. More challenging and time consuming was the addition of a world vector map listing all the countries I have visited. I used the jQuery module JQVmap which provided all the core functionality but was difficult to fit and get at the correct scale. This website will always be a work in progress, so keep checking in to stay updated!

August 2020 - December 2020

In the fall of 2020, I took ECE 5970 Machine Learning with Biomedical Data taught by Mert Sabuncu. For our final project my team implemented a convolutional neural network for segmenting low-grade glioma in brain MRIs. I was the team leader, delegating work and overseeing the overall design of our implementation in PyTorch. We ended up using a standard UNet and focused on different preprocessing and data augmentation techniques to improve accuracy. I also experimented with skull stripping to try and improve our model’s performance which had mixed results. For this project we used Google Colab to train our model using their free GPUs.

To view our Google Colab notebook check the link below on the left. A link to our final report for the project is also included.

June 2020 - August 2020

In the summer of 2020, I worked as an intern at Maculogix, a biomedical startup. The company specializes in building equipment to detect age-related macular degeneration. The product I worked with was the AdaptDx Pro, a portable VR headset-looking device that could detect AMD within 15 minutes. When I arrived at the company there was a production bottleneck with the final calibration of the device, a process that took about 2 hours and 15 minutes. I was tasked with cutting down the time for a 40-minute section of this calibration process, a task that had to be repeated for each side of the device totaling 80 minutes. I was given instructions to implement a modified binary search that at best, would have cut down the time in half. However, after running a series of experiments, collecting data, and analyzing it, I realized that by using a modified interpolation search, that I could not only decrease the time by more than half but also increase the accuracy. In the end, I was able to cut the section down to 10 minutes for a total reduction of 80 minutes to only 20. After realizing how significant of a change had been made, my boss tasked me with cutting down the time for the entire calibration pipeline. Using computer vision algorithms and a FLIR camera, I was able to automate several steps in the calibration process that were originally performed manually, such as alignment of internal cameras and lights. In the end I was able to reduce the time for a complete calibration from 2 hour 15 minutes to only 35 minutes. This eliminated the production bottleneck and allowed the company to cut costs by not having to hire as many calibration techs as they had pla/beenned on.

While the calibration software was my main contribution, there were a couple other things I worked on during my internship. I wrote a 40-page ISO 9000 compliant verification and validation document for my calibration software. I also discovered during my experiments with the device, that some of the sensors used for calibration were incapable of measuring light at the microlux level as accurately as my algorithms needed. After bringing this up with my boss, I was placed in charge of collecting data specifically for our sensors and sharing this with company we purchased them from.

August 2020 - December 2020

In the spring of 2020, several members of Cornell Biomedical Device broke off to create a startup, Vita Innovations. Our product, VitalMask, is a smart mask that measures patient vitals and is designed for ED waiting rooms. The idea was born during the beginning of the Covid-19 as a response to ED overcrowding. I worked on the website (see link below) and the desktop application which was written in Python using Kivy. The application allows for users to monitor multiple masks simultaneously through Bluetooth connections. The application allows for custom thresholds to be given to each mask’s vitals that sound an alert when they are exceeded.

The startup is part of several accelerators and won 15k in seed funding after winning the Business Today’s 2020 Impact Challenge. For more information about VitalMask, check out the website link tomorrow.

August 2020 - December 2020

BruxFree is a bruxism monitoring system that alerts users when they are having an episode. The idea is the user can wear an external device with sensors that detect when the muscles of the jaw are clenching excessively. I pitched this idea in the fall of 2019 to my engineering team (Cornell Biomedical Device) and we worked on it in the Spring of 2020 for the Engineering Innovation competition, for which we placed as finalist. When we started building device, our product development team split into hardware and software. I led the latter in the creation of an android app that could be hooked up to our device and collect data and send notifications to users. You can find more information about the device on the engineering team’s website linked below.

August 2020 - December 2020

My main project for the Schaffer-Nishimura lab was the development of a MATLAB pipeline for pre-processing multiphoton acquisitions. Named the Multiphoton Image Processing Stream (MIPS as an homage to the processor architecture), this project was a yearlong endeavor and resulted in a program that serves as a standard for the lab. Members of the lab use the program to perform basic operations such as channel splitting, normalization, and comb correction among more. There were also more advanced features, such as unmixing and the use of custom TIFF tags to document operations performed on acquisitions. The code performs much faster than MATLABs standard operations as reading and writing images is performed through the ScanImage library.

August 2019 - December 2019

In the fall of 2019, I took CS 3110 Data Structures and Functional Program taught by Michael Clarkson (one of the best professors at Cornell). As our final project, my team decided to implement a database management system in oCaml using SQL as our standard. We implemented multiple SQL queries such as INSERT, DELETE, SELECT, WHERE, LIKE, ORDER BY and SELECT IN. We also added some of our own custom queries to create and delete databases. As my contribution to the project, I was in charge of the REPL code, the display of tables in terminal, and helped with implementation of some of the queries. Our professor was impressed enough that he emailed our team inviting us to demo the project to him personally.

Below on the left is a link to the original project charter.

February 2019 - May 2019

In the spring of 2019, I took CS 3410 Computer Systems Organization and Programming taught by Hakim Weatherspoon. As part of a 3-month long project, we built a mini RISCV processor using Logisim. Our design was heavily based upon the Computer Organization and Design RISC-V Edition.

Below is an image showing the entire schematic for our processor. Due to the scale of this project, many of the final components (such as the ALUs) are abstractions of what we built earlier in the semester. A pdf detailing the entire breakdown of the project is attached in the bottom left.

June 2018 - August 2018

In the summer of 2018, I returned to university after two gap years abroad. To readjust to the rigor of Cornell, I participated in the Engineer Summer Math Institute (ESMI) where I was introduced to student research. For my team’s research project, we were interested in analyzing different soccer formations used in the FIFA World Cup occurring at the time. In Python, we wrote a soccer simulator that incorporated player movement, passing, and shooting. The simulator allowed for the user to input different team formations and evaluate their performance with metrics such as scoring and possession. At the end of the summer, we had the opportunity to present our research at a symposium at Cornell.

Below is the poster we made for the symposium. You can download the pdf in the bottom left.

2017 - 2018

I’ve always been interested in Instructables competitions and how there is the opportunity to not only build something cool but share with others how you did it. As of today, I’m a 7-time Instructables winner (1 grand, 3 first, and 3 second) and my tutorials have over half a million views. Most of my winning submissions have been for leatherwork, something I got into while travelling through Ecuador in 2017.

I remember there was a time when I felt like it was impossible to win an Instructables competition. Now that I have a couple under my belt, I think I’ve realized the importance of perceived quality and presentation. Most Instructables that are featured or win, are also the ones that are presented or look the best. High quality images and a nice thumbnail are of the highest importance when trying to gain votes. Even if what you’re making isn’t that special or innovative, if you can present it in a way that is appealing then there’s a good chance you’ll be in the running for a competition. Another thing that helped my submissions was the time I spent making companion YouTube tutorials. These helped engage the audience on a different modality and provide more concrete visual aids.