Deep Learning Specialization

Sequence Models

Convolutional Neural Networks

Improving Deep Neural Networks: Hyperparameter tuning, Regularization and Optimization

Neural Networks and Deep Learning

Structuring Machine Learning Projects

Machine Learning Specialization

Algorithms on Graphs

Data Structures

Algorithmic Toolbox

Machine Learning: Clustering & Retrieval

Programming for Everybody (Getting Started with Python)

Python Data Structures

Using Databases with Python

Using Python to Access Web Data

Machine Learning: Classification

Machine Learning: Regression

Machine Learning

Machine Learning Foundations: A Case Study Approach

The Complete Java Developer

Co-founder and Mentor

•	Resolved issues involving documentation, API changes, bug fixes & memory enhancements.
•	Reviewed pull requests by other contributors.

Open Source Developer - scikit-learn

City Ambassador

•	Applied Computer Science with Android - An initiative by Google.
•	Gave an introduction to the fundamental concepts of Java, Object Oriented Programming and Data Structures including HashMap and HashSet to get started with Android Development.
•	Conducted 2 workshops helping the students build their first Android application through the concept of pair programming.

Workshop Facilitator

Alok is an initiative by Bhavan's Gangabux Kanoria Vidyamandir, Kolkata to educate the less privileged students by providing the necessary materials and the environment to study along with the students of the school volunteering as their teachers.

Teacher

Climate Change: Learning for Action

Bachelor of Technology (B.Tech.)

High School

Working with the tech team on developing and improving our tree counting and biomass estimation algorithms.

Advisor

• Created a 6‐month Product Fellowship Program, aimed at enabling budding engineers and designers to build product for teaching thousands of students from low‐income backgrounds across India while coaching them on the relevant skills. Owned the end‐to‐end process from inception to recruiting, program design, onboarding, mentorship, training and maintaining production‐quality code. Meet the fellows here.
• Built an open‐source web‐based quizzing engine based on a variant of the IMS QTI Specification 2.2 to be used as an homework platform for Avanti’s Live Classes as well as an assessment platform for mock JEE/NEET tests, using VueJS + Typescript and FastAPI + MongoDB.
• Mentored the creation of Portal, a tool for creating link‐based, custom authentication layers to help identify student profiles across different learning sources. The tool was used for 8k+ users across 400+ schools.

Senior Product Engineer

• Plio is an open‐source tool that empowers creators to convert their videos into interactive lessons by
overlaying questions between the video and gather rich data to understand how their students are learning.
• Plio was used by Avanti Fellows ‐ an 11‐year old education non‐profit, to teach more than 40,000 students from low‐income backgrounds during the COVID‐19 lockdown where we found a 5 times increase in completion rate compared to learning from passive YouTube videos. Our position paper summarizing this work was accepted into the Designing for New Forms of Vulnerability workshop in CHI 2021. Plio has also been used by Peepul to conduct training for 5000+ teachers.
• Designed and developed the microservices for the backend (Django, PostgreSQL) and the frontend (VueJS, Tailwind CSS), REST APIs (Django Rest Framework), CI/CD pipelines (Github Actions, Docker, AWS ECS), Unit Testing (Jest, unittest, Cypress, Codecov), data pipelines (Lambda, BigQuery, AWS RDS), with a focus on makin

Builder

• Developed an early pest-infestation detection framework using real-world images collected and annotated by our team, using the state-of-the-art deep learning based object detection algorithms like SSD, RefineDet, etc. to achieve < 5% false alarm and missed alarm rates. The subsequent models have been compressed using methods like iterative pruning and quantization, reducing the model size from >250 MB to 5MB with almost no drop in performance, for offline inference on a smartphone. Our paper has been published in KDD 2020 and we received a $2M grant for the project as part of the Google AI Impact Challenge offered to only 20 teams worldwide, becoming the only team from India to do so.

• Built an end-to-end pipeline to identify low-birth weight babies by learning to estimate weight directly from smartphone videos only, using a combination of a learned deformable model that captures the latent subspace of realistic baby shapes, and domain adaptation from synthetic to real data. Our te

Research Fellow

• Worked on the problem of change detection where the inputs are the images of a place at two different timestamps and the expected output is the change that happened in the time duration. A lot of factors make this problem challenging  like misalignment of images, lack of high quality labeled data and even factors like seasonal changes / shadows that might make the images look very different even though not much has changed. Change detection finds a lot of useful applications in defense, studying the effect of climate change, etc.

• Designed the pipeline by framing this as a semantic segmentation problem and used various architectures like U-Net, V-Net and DeepLabv3+ combined with trying out different loss functions, activation functions like PReLU and Maxout,  variants of the original architectures,  optimizers like SGD & Adam along with various regularization techniques like Data Augmentation (geometrical/color) to monitor and improve the segmentation map to obtain a pixel-wise acc

Deep Learning Intern

• Developed the pipeline for Semantic Segmentation on thermographic images implementing various Convolutional Neural Networks architectures like U-Net, V-Net, Convolutional Sliding Window and a multi-channel cascaded architecture.

• All the scripts were written in Python, using Keras as the framework, with TensorFlow backend. Wrote various custom layers and experimented with different metrics like weighted cross-entropy, dice coefficient, etc. to improve the specificity and sensitivity of the model.

• Utilized Bayesian Optimization for performing Hyperparameter tuning, using a wrapper for Keras called Hyperas.

Our paper has been accepted into the Quantitative InfraRed Thermography Conference, 2018.

Research Intern - Deep Learning

•	Developed a real-time car parking system model applying a Convolutional Neural Network (VGGNet) on CCTV camera images using the PKLot dataset for training and achieved an accuracy of almost 94%.
•	Proposed a novel algorithm for internal navigation using Li-Fi with a graph-based approach
and preventing car-theft using Machine Learning & Image Processing.

Software Engineering Intern

•     I co-founded Scholar Mate during my sophomore year to provide an efficient platform for effective communication between teachers, parents and students. Led a team of 5 people and launched in 1 school with 50+ daily active users and 100+ installs in the first week.
•	Engineered the front-end of the Android Application (Java, XML) as well as the server side code (PHP, MySQL) using AWS as the cloud.

Co-founder

Tech Lead

Tech Lead @WisdomCircle (marketplace for retirees to find meaningful things to do post-retirement)

Know about Aman Dalmia (hiring)'s work experience, education, and much more. Aman Dalmia (hiring)