Last Thursday, some of the top machine learning engineering talent in Asia spent the better part of their day in a neck-and-neck competition at WorkFusion’s Ascend conference in Bangalore, India.
The task was to build a machine learning model that extracts information from invoices using the WorkFusion ML SDK. The challenge had to be finished within 7.5 hours, a very short timeframe for such a complex task.
This was our second hackathon, after the premiere event at Ascend NYC in May where EPAM walked away with first prize. Another repeat participant was Capgemini, whose team was highly driven. The team spent almost a week prepping by learning the SDK architecture inside out. “We gave some of our best ML engineers time off from their ongoing assignments so they could participate,” says RPA architect and machine learning engineer, Amol Kankane. “My experience from the WF Labs also helped a lot. It also made it easy for me to reach out to the WorkFusion team whenever we had any queries and doubts.”
The competition started at 9 am and finished at 4:30 pm. Teams representing veteran participants Capgemini, Cognizant and EPAM were fighting along with first timers Tech Mahindra, Infosys, DXC, Accenture, and Fujitsu.
The teams were presented with a dataset that was split into three parts: training, validation, and test. Training is the actual dataset used to train the model, validation is used to fine-tune the given model, and test provides the gold standard used to evaluate the model. It’s only used once the model is completely trained (using the training and validation sets). Raising the difficulty level was the realistic aspect that some of the datasets contained errors, such as missing fields, OCR errors and duplicate information.
After an intense competition, where the top position kept fluctuating among the Capgemini, Accenture, DXC and EPAM teams, Capgemini’s meticulous preparation appeared to pay off. The team nabbed the grand prize of $1,500 with a final average score of 0.891709, which is a very impressive achievement considering the time crunch.
But even though the competition was fierce, the spirit among the contestants was friendly.
It was a very nice event and there was more cooperation than competition. It also helped us to get more confident about WorkFusion’s machine learning.
Amol Kankane expressed similar sentiments: “Overall, it was an amazing experience, which gave us motivation to learn the new ML capabilities in a short time and demonstrate how well we can perform as a team. Winning the hackathon definitely helps us gain more confidence in building robust solutions for our clients.”
Congratulations to all on a job well done!
Also published on Medium.