Flashparse
The Challenge
This was actually a project we worked on as part of our previous company, Gathr. The three founders of PressW built this solution together and it dramatically changed the way our previous product worked and became one of our core IPs as part of our acquisition.
At Gathr we were looking to reinvent the way social events were planned from the ground up. We believed that filling out forms (think Facebook Events) were the way of the past, and wanted to build a system that was able to take in just a sentence like “Dinner tonight at 9pm at Red Lobster with Tarun, Liam, and Bryson”, and in real-time extract all the relevant event details and setup the plan. The goal was to be able to create a plan faster than you could send a tweet.
The Solution
FlashParse is a state of the art Natural Language Processing engine that combines the ability to reliably extract event details such as title, date, time, participants, and location from blocks of unstructured text by leveraging the latest Natural Language Processing techniques.
We built Flashparse to also integrate with other services so when you typed “Red Lobster” it was even able to figure out the most relevant Red Lobster location and tag it on Google Maps for the plan creation.
In addition we built a clean, highly functional UI to sit on top of the model, which allowed users to see event detail extraction in real time as they input their event as a freeform string (similar to a text message). This UI was crucial as it allowed the user to accept or decline the recommendations inline so event creation was a near instant process, with the added benefit of allowing us to gain additional training data to customize each user’s experience from this yes/no paradigm.
The entire web of models were each hosted on Google Cloud, and architected to be horizontally scalable as individual entities. This allowed us to achieve sub 1 second processing time per request and handle loads greater than 10,000 concurrent requests per second.
The Outcome
Flashparse was a huge success point for our last company. It dropped our average plan creation time from 2.6 minutes to less than 30 seconds consistently, and it significantly improved the number of plans being created. The stellar UI and unique creation engine that we constructed help garner attention and ultimately became a cornerstone for our acquisition.
Connections
10,000
Per Second
