ate Sawant

ate Sawant

Software Engineer
and Photographer

My name is Nate Sawant, a software engineer based in Boston, MA. I’m about to graduate from Northeastern University with a BS in Computer Science, concentrating in Artificial Intelligence.

I love working on a wide range of projects, from computer vision to game development to web development. I’ve also had the privilege of working with so many talented people and on different projects through Generate, Northeastern’s student-led product development studio. You see some of them below.

Outside of my academics and professional life (which you can also see below), I enjoy spending my time outside either climbing rocks and shooting photos of the great outdoors.

Experience

More to come soon!

Significantly improved system performance by doubling download speeds through multi-threading and reducing memory usage by 20% during Named Entity Recognition (NER) training. Conducted an in-depth analysis of NER categorical changes across millions of items using Python and ontology, ensuring consistent data stability. Additionally, streamlined open-source compliance processes by automating license report generation, resulting in significant time savings for each Maven project.

Python
MongoDB
Java
Spring Framework

Successfully streamlined development processes by leading the creation of a cutting-edge shared code snippet tool for Visual Studio Code and Jupyter Lab using TypeScript, which enhanced collaboration and efficiency across teams. Orchestrated comprehensive ETL processes on over 10 websites, scraping and consolidating data points for more than 65,000 items, while employing Python to seamlessly integrate this data into SQL databases. Additionally, developed an insightful Tableau dashboard to track usage patterns across 100+ datasets, empowering decision-makers with critical insights to drive data-informed strategies and enhance operational effectiveness.

Python
SQL
TypeScript
Selenium

Developed engaging two-player gameplay mechanics utilizing the Unity 3D engine, integrated with OpenXR and Photon Unity Networking to create a seamless multiplayer experience. This involved designing and implementing interactive features that enhanced player collaboration and cooperation. Additionally, conducted in-depth analysis of participant behavior by capturing eye-tracking data at 60Hz using the Vive Pro headset. This data collection allowed for a detailed examination of focal points within the virtual environment, providing valuable insights into user attention and engagement patterns.

C#
Unity
OpenXR

Projects

You are currently here. Designed in Figma and implemented with React. Hope your web crawler doesn't get stuck in an infinite loop going to the GitHub repository and back here. I wanted this to be a place to showcase all of my work, from professional to personal projects to games and photographs.

Figma
Astro
React
TypeScript
Tailwind

Snapper is a gamified social platform for divers and marine enthusiasts to share, find, identify, and learn more about the marine life around them and the world. Users are able to share photos, descriptions and 'collect' fish they’ve spotted during their dives, allowing others to quickly see what fish are being spotted in an area or where a certain fish is being spotted. Gamification comes from the Pokedex-esque way of collecting fish and badges as rewards.

Python
MongoDB
Java
Spring Framework

DevDB serves as a tool to rapidly deploy development databases for Generate teams. Leveraging the beauty of the Prisma schema, users can deploy and setup databases with either the CLI or website, allowing them to choose whatever branch they are using. The databases are also automatically updated when the schema is updated using GitHub webhooks. Data can also automatically be generated using LLMs and the schema, so manual sample data generation is a thing of the past. All of this is deployed on AWS.

PostgreSQL
AWS
TypeScript
tRPC
Prisma ORM

One of the biggest clubs at Northeastern is the Northeastern University Hus-skiers and Outing Club, simply known as NUHOC. To help people to enjoy the great outdoors, NUHOC rents out gear (tents, sleeping bags, etc.), but is done through a very manual process with spreadsheets. To help everyone out, I created the new NUHOC site where members can sign in through Google SSO, check their membership status, reserve gear and see what they gear they've rented out. It also serves as a dashboard for eboard members to maintain the inventory and manage members.

PostgreSQL
TypeScript
React
Prisma ORM
tRPC
Figma

Tubender is a CNC-powered tube bending machine. Capable of bending tubes are precise angles and lengths, it brings the cost down and brings the ability to bend tubes to the common person. The Generate team developed the software to interface with the machine, allowing a user to upload a CAD file of the tube and calculates the machine instructions to execute the bend. The C++ backend uses the OpenCASCADE libray to read the CAD files and outputs the instruction. A desktop application was created using React, Electron, and backend.

C++
Electron
TypeScript
React

Chess has always been there in my life from a young age, I may not be the greatest but its one of my pasttimes, especially online chess. The essentially perfect decisions of computers in chess (and games in general) is one of the things that inspired me to pursue computer science and AI. Built in Golang for better performance, a game tree with alpha-beta pruning was used for the general structure. Different metrics and heuristics were used to evaluate the position of the board, which allowed the computer to determine what the next best move was.

Golang

Voxeti is a peer-to-peer 3D printing platform for those who want to print things can find people who have 3D printers. Users are able to upload the model files, select the filament, colors, and scale to submit the job. Producers are then able to browse the offered jobs, choosing to accept or reject, and then fulfil the order to be shipped off to the customer. To estimate the costs for the jobs, the open-source C++ slicer CuraEngine was modified to estimate the print time and filament usage, without producing any other files.

Golang
MongoDB
C++
React
TypeScript

Reinforcement learning in robotics is becoming more popular, as robots' behaviors can be trained and fine-tuned in simulators rather than in real life. It beats having a human try to manually optimize the behavior. As a way to investigate this field, a simple agent with an array of raycasts to simulate the LIDAR of a real robot was used as a surrogate. Trained and evaluated in a randomly generated continuous maze environment, the goal was to reach a certain target object. The training was over 1,000 iterations and was able to reach the target without hitting the walls around 75% of the time.

PyTorch
C#
Unity