8090's Top Coder Challenge

4 minute read

Original LinkedIn Post

On Friday night, I saw a public post on Twitter/X by Chamath Palihapitiya announcing an open Top Coder Challenge hosted by his new company, 8090 Solutions. Anyone could join. The challenge would take place the very next day, last just 8 hours, and involve reverse engineering a black box legacy system using only historical data and a few employee interviews.

I decided to jump in.

By the end of the day, I was honored to place 7th out of 425 engineers. You can checkout the the leaderboard HERE and checkout the code for this challenge HERE. But, I will not lie, I was honestly just hoping to finish something within that short period of time, so making it onto the leaderboard was a surprise and a huge personal win for me.

The challenge was solo, and the goal was to replicate a 60 year old black box travel reimbursement system that had no source code and no documentation. We were given a few artifacts including a product brief, transcripts of employee interviews, and a public dataset containing 1,000 historical examples of inputs and expected outputs. From that, I had to infer the business logic behind how reimbursement amounts were calculated and implement a modern version that could produce the same results as closely as possible. Submissions were evaluated on a separate hidden dataset that contained 5,000 test cases instead of the original 1,000. This larger private set is what ultimately determined your final score and ranking. The scoring system rewarded accuracy, where a lower score meant your solution more closely matched the hidden behavior of the original system.

To tackle the uncertainty and patterns in the data, I used classical machine learning techniques alongside basic heuristics and programmatic logic. It was a careful blend of data analysis, feature modeling, and rule approximation based on imperfect clues.

Here was my eval score for the public 1,000 dataset:

✅ Evaluation Summary
------------------------
  Total cases        : 1000
  Exact matches (<$0.01): 0
  Close matches (<$1.00): 17
  Average error      : $31.15
  Score              : 3214.93

Developing a solution to such a challenge in 8 hours would have been nearly impossible without the help of AI-powered tools that made it easier to rapidly explore, integrate, and test ideas.

It felt like software archaeology combined with a live coding sprint. Easily one of the most intense and rewarding technical challenges I have done.

Thank you to Chamath Palihapitiya and Arjun Krishna for organizing such a creative and inspiring challenge.

🔗 Links: