Aug 2022 – Oct 2024 · 2 yrs 2 mos · Amsterdam, North Holland, Netherlands

• GreenSwapp AI is a data platform for food services, retailers & brands to harmonize product data & track Scope 3 emissions per product or recipe for their entire range in real time. Our data helps customers (1) save time by automating ESG reporting (2) meet internal climate goals quicker by identifying lower-impact SKUs to procure and (3) increase consumer awareness by communicating impacts on shelf labels or packaging.
• Customers integrate our data directly using our APIs or via channel partners (like Google Cloud) where it is pre-integrated. We are 1 of 14 Google Startups for Sustainable Development and are a Tier 1 Startup in their Climate Action Challenge - both of which support us with financial and technical resources.

Dec 2019 – Apr 2022 · 2 yrs 4 mos · San Francisco Bay Area

• Peloton for Kitchen :)

Jan 2017 – Jan 2019 · 2 yrs · Gurugram, Haryana, India

• IoT Platform enabling interoperability across communication protocols

Jul 2016 – Mar 2017 · 8 mos · Greater New York City Area

• Worked at The Sage Project, Now Pinto. Pinto is working on nutrition data and aims to provide personalized nutrition.
• Created a pre-processing pipeline on real-world multi-type unstructured data, Used multiple scaling and dimensionality reduction methods to efficiently reduce required dimensions for further processing.
• Used unsupervised learning methods such as Hierarchical density-based spatial clustering of applications with noise (HDBScan) and hierarchical agglomerative clustering for hierarchical categorization of supermarket products. Worked on dendrogram visualization for visual analysis of the result.
• Used multiple ensemble learning methods such as AdaBoost, Random Forest for testing and building a general classifier based on categorized data. Used different methods for bagging, boosting and parameter tuning to choose the combination of models. Used Grid Search and Random Search for parameters and model selection.
• Deployed & Automated the complete pipeline with Python API.

Apr 2014 – May 2014 · 1 mo · Remote

• Worked as an open-source contributor at the PaGMO project, Advanced Concepts Team, European Space Agency. It was short but pivotal for me to learn important things and get entry to this field of study. It helped me in the process of transitioning from a Python Developer to C++ Algorithms Programmer by researching, understanding and implementing a state-of-art genetic algorithm and test problems. This experience led to more exploration and research in this field.
• The underlying concept and similarity with real-world problems intrigued me, as most real-world complex problems consist of multiple objectives and multiple types of parameters. Quite ironic that we also use algorithms inspired by the evolutionary process to solve these optimization problems. I'll be happy to return and contribute more in the future.
• PaGMO (Parallel Global Multiobjective Optimizer) is developed to solve constrained, unconstrained, single objective, multiple objective, continuous and integer optimization problems, stochastic and deterministic problems, as well as to perform research on novel algorithms and paradigms and easily compare them to the state of the art implementations of established ones.
• Implemented Non-Dominated Sorting Genetic Algorithm (NSGA-2) in C++ so It can be used in Integer Multi-objective problems. (NSGA-2 by K. Deb)
• Implemented test problems to test the algorithm performance
• Wrote test cases and documentation

May 2012 – Aug 2012 · 3 mos

• This was part of a Google Summer of Code program. It was under Interface Ecology Lab @ Texas A&M University. The Interface Ecology Lab investigates the future of human expression, focusing on creativity, play, participation, and learning. For this, they have introduced the concept of information composition. An information composition is a spatially arranged set of rich bookmarks that visualizes concepts and ideas. The inefficiency of text, audio, and video require us to search for a better medium to express concepts, ideas, and connections between them. However, we already have different graphical and matrices based structures to show these connections but they're limited in terms of data types and they're not accessible to the masses.
• I designed and Implemented a social network for Information compositions. The social element is important because connecting people has always been an integral part of sharing ideas, concepts, and connections. A concept/idea/connection/theory/meme can only improve if it can travel through masses.
• Designed and Implemented a complete social network (Now Idea Mache at Ecology Lab, TAMU) with social elements of sharing, editing, uploading information compositions.
• Implemented features such as friends, messaging, threaded comments, a tag cloud, administration, etc.
• Implemented indexing and search using Xapian
• Designed and Implemented REST based python API to add, delete, update, download operations on Information composition

Jun 2011 – Jan 2012 · 7 mos

• Research Intern (Remote), Human Development Lab, Project ROWDI
• ROWDI - Reading Out World Digitally
• ROWDI project was aimed at helping middle school students to maximize their learning curve in literature studies using the interactive medium. We designed a total of two motion sensing games using Kinect and Unity3D. The first game was based on "The Evolution of Calpurnia Tate" by Jacqueline Kelly, the other one was based on abduction story from "The Door in the Lake" by Nancy Butts. The assumption was if the interface is interactive, immersive and literature lessons are taught through a game, the learning, retention, and attention can be improved.
• Unity3D is used for gameplay and physics.
• Used OpenNI & SITE to integrate Unit3D gameplay and scenes to recognize and send depth, gesture and location data to Mircosoft Kinect Middleware
• Deployment was done at McPherson Middle School, Kansas for research in improvement in learning.