Portfolio

Our client (a YCombinator-backed startup) wanted to train their own base model to generate images from user-entered prompts (much like MidJourney or DALL-E). They hired us to prepare the dataset for model training.

In order to tackle this challenge, we had to process 100 Terabytes of images (100 million images) and prepare them for training. This required significant cloud infrastructure and performance optimizations. We used Postgres on AWS RDS and AWS SQS (Simple Queue Service) for workload distribution and task queuing. We built a pipeline to handle image processing that divided the data into manageable batches for efficient processing. The project focused on building scalable systems and solving the challenges of parallel data processing across distributed servers. We monitored the entire infrastructure with Datadog to ensure we had strong viability into the results and job processing.

Read more about the project on our blog →

Welcome SearchNEU, an alternative to Northeastern University's cumbersome course catalog system, Banner. Founded by Ryan Hughes in 2017 to simplify course registration, SearchNEU addresses Banner's limitations by offering a mobile-friendly interface, faster search capabilities, and notifications for open seats. The site is built on Amazon Web Services, stores data in Postgres on RDS, and the frontend is built with Typescript and React. After gaining widespread popularity, the project was passed to Sandbox, a student-led consultancy, which continued to enhance the app's functionality. Over the years, the app has seen numerous updates, including transitioning to different infrastructure, adding support for additional colleges, and improving responsiveness. Today, SearchNEU remains an essential tool for Northeastern students and faculty, vastly improving course registration and access to class information.

Read more about the project on our blog →

We stepped in to help a client who experienced a critical data breach. An attacker deleted our client's MongoDB database and demanded a ransom to be paid in Bitcoin. The business, which had been relying on infrequent manual backups, faced significant downtime and data loss. We worked with the client to recover their data and outlined the importance of automated backups, limiting access for remote engineers, implementing a staging environment, and refining the hiring process. These changes helped the business enhance its security infrastructure and operational practices to prevent future incidents.

Read more about the project on our blog →

Introducing Eames, our new Mac app designed to keep you on time for your meetings. We've all been there—caught up in work and suddenly realizing we're late for a call. Eames is here to solve that problem by automatically launching your meeting windows on time. It integrates seamlessly with Google Calendar, Zoom, and Google Meet, and will soon support more platforms like Microsoft Teams and WebEx. With Eames, you'll never miss the start of a meeting again. Download it today and stay punctual effortlessly!

Read more and download Eames on our blog →

We had a client who faced unauthorized and significant data scraping on their website. Unfortunately, they didn't have the budget to purchase expensive Bot Management Solutions so they hired us to shut down the scrapers and keep their data secure.

First, we implemented web request logging with SumoLogic to fully understand the scrapers. This additional logging revealed a high volume of suspicious requests. Then, we deployed a custom in-house solution using AWS Web Application Firewall (WAF). Our solution combined reverse DNS lookups and cookie tracking to distinguish legitimate traffic from malicious bots. This approach reduced bot traffic significantly, successfully protected the client's data, and stayed within their budget. The client continues to use our custom management solution solution to this day and is amazed how many bots it blocks.

Read more about the project on our blog →

Efficiently fashion maintainable leadership via enterprise users. Globally myocardinate sticky technologies for focused intellectual capital. Authoritatively reintermediate extensible processes whereas enabled core competencies.

Conveniently reinvent cross-unit materials before inexpensive partnerships. Phosfluorescently brand distributed relationships whereas proactive partnerships.

Completely extend front-end total.