FeedBee
Recently, I've been subscribing to more and more newsletters. It's a great way to learn about
cool new products,
ideas and all kinds
of interesting things! Newsletters are meant to spark thoughts and discussion, however, it's currently a one way street. Writers share their content
but often have very little knowledge of what their readers think of it. Me and 4 of my friends wanted to change that! So we decided to build
FeedBee to help newsletter writers get better feedback.
We wanted to make the process more direct for readers by allowing them to voice their thoughts through links in the newsletters themselves!
Their click directly records their 'vote' in our database and then sends them to a form where they can provide more personal insights. We also wanted to
present all this data to writers in an organized manner with engaging charts and dashboards.
On the team, I interviewed newsletter writers to understand their needs and also did some design stuff but primarily worked on developing the software. We launched on
product hunt and got 7 customers on our free trial! :)
I enjoyed working on awesome new features and improving FeedBee with my incredible team!
Human-Style Doodle Generation
I've always found animation, especially hand-drawn animation and doodles to be so enchanting because they have the power to
convey stories and worlds that we simply cannot capture through another medium. This is why for the final project
of my Deep Learning class, I partnered with another Master's student to explore the generation of human-style doodles. I
came across the
Google Quickdraw game, found it to be very amusing and was excited to do something with its
publicly available data.
Usually image generation in machine learning makes use of pixel based representations of images, but past work in drawing generation
in particular has taken a vector based approach where images are represented as sequences of points instead given the dynamic nature
of drawings. We found this to be a fascinating approach and explored different existing models that used it. Most of them were highly complex even
though they produced good results. This motivated us to develop and test 4 different simple architectures that incorporate GANs
and RNNs with vector based representations of sketches.
CSF Global Partnership
I was the project lead (along with the amazing Julia Ring) for
CSF Global Partnership, a student-run project team.
Our team formed when Dr. Tasneem Karim, a researcher at an NGO,
CSF Global reached out to collaborate on a project
that aims to empower children with Cerebral Palsy in Bangladesh. We conducted a
needs assessment in Sirajganj, Bangladesh in Summer 2019 to better understand the day to
day challenges faced by children with Cerebral Palsy and their caregivers. We met with Bangladeshi nutrition
experts such as Dr. Khurshid Talukder and Moushumi Khan, who is a UofM alumna and runs the NGO Foundation for Charitable
Activities in Bangladesh (FCAB), before and during our needs assessment trip to gain more insights into the challenges and how we can make our future intervention effective.
Several families mentioned difficulty feeding their child and that their child was unable to feed themselves.
We also observed mothers feeding their children during therapy sessions, instead of the child working to feed
themselves. Children with Cerebral palsy also have a higher incidence of malnutrition because they have difficulty
chewing and swallowing.
This is why our team decided to focus on alleviating the childrens’ feeding and nutritional problems. The education subteam is working to create an information booklet that allows caregivers to easily access posture tips,
ways to modify the texture of food and recipes for affordable food that is best suited for children
who have Cerebral palsy. Whereas the assistive device subteam is working to create an assistive plate and glove
that makes it easier for children with Cerebral palsy to feed independently.
The Spirelephant
My teammates and I created the Spirelephant as part of EECS 495: Software for Access. After surgery, the alveoli
in the lungs of patients close up, which is why doctors recommend that they do breathing exercises using a device
called a spirometer. But patients often do not do these exercises because it feels too monotonous and doctors do not
have any way of checking up on them. This problem is worse for young children who may get bored easily. Enter: the Spirelephant!
It uses the patients' breaths as input to the game to control the on screen character Elly the elephant who completes various
tasks to help out her animal friends. I had a lot of fun creating the art for the game!