Competitive Crowdsourcing to Scale your AI Journey

AI based solutions are starting to go mainstream. Enterprises are beginning to ascend in the AI maturity curve – from experimentation to industrialization. With proliferation of digitally available data and hyperscale computing, advances in the AI related technology are happening swiftly. However, one of the key challenges there is, and will remain, for the foreseeable future is the availability of good talent – people with adaptability and ability to upskill quickly with these leading edge technologies. There are many approaches at play for organizations to tap into the global talent pool, even so we came across an interesting strategy by participating i.e. competitive crowdsourcing events. Several medium and large enterprises are tapping effectively into this approach to help them gain access to good talent as well as secure a faster solution to the wicked problems they are facing.

This blog discusses our learnings and why as a services or product company there is a greater good in getting involved in such challenges.

The events we participated in attracted talent and teams from across the globe. Engaging in such events serves as an eye opener into how organizations are willing to share their data with a public and global pool to avail some of the best algorithms and solutions in a short time frame. This also provides some insights into potentially how organizations can use innovative procurement approaches to secure products and services in a democratized manner, thus creating a win-win situation for all players in this emerging ecosystem.

Our AI & Automation Services team recently partook in various public AI challenges, created by organizations in Australia. The challenges were:

1.        How can AI help detect underperformance to then improve output of loaders used in underground mining?

2.        Based on assessment of past incidents at an energy company, predict when an OH&S incident may occur.

3.        Hackathon between telco client and partner organizations to understand the best use of AI technologies in their enterprise.

4.        How can AI be effectively used to solve challenges in the public sector – issues that cover local, state or federal government areas of interest?

For the Telco hackathon, there were no restrictions on the use cases we could address. We effectively leveraged Infosys’ own AI platform NIA, to create an empathy virtual bot that can sense the sentiment or mood of the user and recommend multiple actions they could take to relieve stress – be it through simple ergonomic changes or even getting some fresh air or a coffee. With the Government hackathon challenge, we leveraged open source technology to demonstrate how we effectively predict bushfires based on data sets that were made available to the team. The government had provided a wide range of local, state, and federal problem statements to pick from with relevant data sets. These two challenges were restricted to participants from the region.

However, for the mining and energy customer challenges, we competed against teams from across all continents wherein the teams were asked to deliver incremental changes to the AI models. The purpose was to compare real time where each team’s propensity scores stood, in a short period of time.

The mining and energy companies worked with Unearthed.Solutions, an event management company that creates and manages such challenges. All new technologies will always go through an experimental stage and clients are capitalizing on the innovations that are happening in pockets, throwing down a good challenge that gets the attention from across the globe.

NASA has been doing something like this for a good part of the last decade through their Space apps challenge.


More about the challenges!

Few characteristics on the challenges we participated in are below. Given the right amount of time and effort, these solutions can be finessed to a highly mature state that can positively influence the outcome for these enterprises.

  Australian Mining Company Western Australia Energy Company
Brief on the Challenge
  • Australian Mining company’s Gold mining site at Cadia uses underground loaders to move ore from the draw points to be crushed as part of their overall mineral processing operation.
  • The challenge is to predict underperformance of underground loaders at Cadia mine so that the mining company can plan preventive maintenance to avoid downtime and increase efficiency of loading process.
  • Western Australia Energy company is an Australian State Government owned Corporation responsible for building, maintaining, and operating an electricity network that connects 2.3 million people.
  • The challenge is to build a model to predict when incidents (accidents) are most likely to occur and provide an explanation of the findings and predictions in a report.
Data sets
  • Data for 12 loaders for 4 months with over 4.6 Mn records
  • Data for 20 Mn+ hours of work data in Western Australia Energy company with 821 incident records.
Our approach to address the problem
  • Our approach to address the challenge involved detailed Exploratory data analysis(EDA) which helped us to identify the importance of time lag parameter and we generated time lag features for each loader as a measure of loader’s historic performance.
  • Post EDA, the team used multitude of models like Random Forest, Gradient Boost and XGBoost to model the performance of underground loaders.
  • As part of Exploratory Data Analysis(EDA), the team did up sampling to balance the data and also generated additional features like “Hour”, “day_of_week”, “day_of_year”, “day_name”, “week” and “Holiday” to model their impact on occurrence of incidents.
  • In Modeling the team tried Interpret ML Models namely Explainable Boosting Classifier and Classification Tree.
Outcome achieved
  • The XGBOOST model built helped predict if a loader will fail one hour later, four hours later and so on and this ensures that the mining company can plan preventive maintenance and avoid an unplanned outage.
  • The Explainable Boosting Classifier model built identified the top reasons for occurrence of incidents and gave insights into time of year when they are mostly likely to occur. Refer Exhibit1 below for details.
Future possibilities
  • The model developed essentially predicts a failure of loader based on past performance and this model can be customized to address similar Engineering solutions.
  • The model developed can be leveraged to predict incidents in other risky job areas and help minimize the occurrence of incidents.
Exhibit1 – Graphs from Western Australia Energy Company challenge



AI-led innovations are here to stay, thanks in part to the technological advances and digital transformations happening all around the world. However, access to good talent will continue to be a challenge. Through innovative approaches such as crowdsourcing, and strong & nimble partners such as Infosys, as well as global talents, there is a solution emerging which can help scale the AI journey for any organization.

Top takeaways from this experience:

·         From an enterprise perspective, you can tap into what you need to solve pressing business challenges through meaningful engagement with the larger AI ecosystem of partners

·         From an AI product/ platform view, there are many such competitions out there that can help you showcase the capabilities of your platform wherein many clients and professional services will be looking to engage with you

·         From  professional services standpoint, given the virtual working mode has been tested and proven through COVID-19, you can easily tap into your global pool and gig pools to be able to compete and potentially set an appointment with a prospective client.



Special credits to Jacob Koshy for co-authoring the blog with me. Jacob engages with clients across industries for opportunities to deliver operational improvements using AI & Automation.

Infosys applied AI is a holistic approach and framework that helps clients define, de-risk and democratize AI within their enterprises. For further details, do reach out to

Author Details

Jeya Prakash Sekhar

Jeya Prakash is a Lead Consultant with the Infosys AI & Automation practice and focuses on AI based approaches to solving client and industry challenges. His areas of interest include Deep learning, NLP and Computer Vision.

Leave a Comment

Your email address will not be published.