My Coding Work Samples
PROJECTS
Disease Detection Machine Learning model for non-invasive diagnostics
- A binary LASSO logistic regression model built using Python and tuned with cross validation to identify similarities and differences in rich molecular spectra gathered by Desorption Electrospray Ionization Mass Spectrometry (DESI-MS) of human sweat and breath. This is useful to detect cancers and other diseases non-invasively, allowing for cheaper diagnostics and faster treatment. I am currently working on this project as a student researcher at the University of Stanford’s Department of Chemistry. (2022-Present)
https://github.com/adimit078/StanfordLASSORegression
Medical Image Classifier Machine Learning model
- A ResNet Convolutional Neural Network built using Python to detect endotracheal tubes, central venous catheters, and chest tubes from radiology scans. The program is developed with 2 convolutional layers, 1 linear layer, 1 flattening layer, 1 pooling layer, 2 ReLu layers, and 1 Softmax layer, using bounding box methodology to annotate digital images with medical devices implanted. (2021)
https://github.com/adimit078/medicalImageClassifier
sFlora - App to Intelligently Detect Plant Diseases
- An iOS application built using Swift that identifies the plant species and treatment for a set of identified diseases from images of leaves. It employs an original classification model created with Apple's Create ML framework with publicly available plant images from Kaggle supplemented with personally gathered images to produce a 54383-image dataset. Further integrations tested with the MobileNetV3 Image Classification Model are used determine which dataset and model pair is best. (2021)
https://github.com/adimit078/sFlora
mYoga - AI-driven Yoga Pose Estimation App
- As part of the mYoga team, I helped develop an AI/ML-driven yoga pose estimation mobile application in iOS that employs the TensorFlow MoveNet Thunder model to detect and correct human joint positions and provide corrections. I was responsible for coding in Swift/Xcode for an iOS app that used the TensorFlow MoveNet Thunder model to identify 17 human joints from input images or live data streams. Taking the joint position data from the Thunder model, we trained a pose estimation algorithm using various publicly available Kaggle image datasets of various yoga positions, to accurately display the error margin between a user’s pose and the correct configuration of the pose identified by the Thunder model in real time. This allows the user with an accurate and simple way to learn yoga by comparing their movements with a correct one and adjusting their stances based on the pose estimation model’s corrective error margins. (2021-2022).
Code shared up on request
Multiple Sclerosis Predictor Machine Learning model
- A logistic regression model built in Python to classify a patient’s state of Multiple Sclerosis disease from differential expression of small microRNA in Peripheral Blood Mononuclear Cells. I utilized publicly available genetic expression data from the GEO2R tool to develop the model, especially for identifying early onset Multiple Sclerosis. I normalizing the data and fine tuned the 50,000 data samples to reduce model complexity through LogFold Change sorting. By importing Pandas, Numpy, and Sklearn, I also implemented a Random Forest Regression model to successfully perform feature extraction on the curated dataset to isolate potential genetic biomarkers. (2020-2022)
https://github.com/adimit078/Multiple-Sclerosis-Machine-Learning
Autonomous Driving Neural Network
- A Python YOLO Convolutional Neural Network built by me and my Inspirit AI team to detect and classify cars and other objects in real time while on the road. Our group compared current YOLO v4 model with Google’s EfficientDet model to identify what makes a better machine learning model, and how to perform result analysis through methods like sensitivity/specificity, AUC, ROC and confusion matrices. (2022)
https://github.com/adimit078/autodrivingCNN
First Tech Challenge Autonomous Driving
- A Java program built by my FTC team and me to detect a certain image and perform tasks for the 2019-2020 FTC season. (2019)
https://github.com/adimit078/Chinmay-Teamcode
SuperSportsStats - A Cool HTML Sports Site
- An HTML website built by my AP CSA team and me that uses JavaScript, CSS and PHP to deliver recent statistics for basketball, tennis, and soccer players’ performances. (2021)
https://github.com/adimit078/Mittal-Samal-Nemani-5
MY PUBLIC BLOGS AND TUTORIALS
“My Journey To Humanity 2.0” blog
- A blog I wrote during my Inspirit AI course that talks about my journey of exploring AI, lessons I’ve learned and advice I’d give for new learners of AI.
“Deep Learning Predicts Neurodegenerative Disorders From Brain Scans” blog
- One of the first articles written on the Neuromatter website, breaking down a recent AI advancement in assessing brain age from electroencephalography data.
“How to Use Deep Learning to Diagnose Neurological Disorders” video tutorial
https://www.youtube.com/watch?v=NlpYoW5TnSc&list=PLQSfJ7QrhDSVLg1KFhssG4PEEs_ZutBdW
- A video I published on Neuromatter’s YouTube channel in which I break down my Multiple Sclerosis ML project and discuss the biology behind the disease, basic AI concepts like linear regression, how to collect and normalize data from the GEO2R tool (this way viewers can reproduce the project and create ML prediction models for any disease, not just neurological!), and setup/code a ML program in python. The Neuromatter channel has over 1000 views.
FUN MINI PROJECTS
iOS App that uses public APIs to report weather
DART application that uses voice-to-text API to organize Zoom transcripts
iOS App that vibrates the phone violently when “SHOCK!!” is pressed (I have a younger brother)
Matlab program to intelligently filter spam emails
iOS “Piggy Bank” app to record savings and spendings
Minor Projects that use Python to create Linear Regression, Bias-Variance Trade Off, Logistic Regression, K Nearest Neighbors, Random Forest Regression, K Means Clustering, PCA, NLP, Neural Networks/Deep Learning