I had a lot of fun couple months ago building a talent insight newsletter:

Then I lost interest doing the same analysis on different companies. So I want to see how I can leverage the crowd intelligence by

  1. Exposing a UI for public access to the Database queries. (Think Dune Analytics for LinkDB)
  2. Public users will be able to make a few queries to the dataset everyday (rate limited) so I can see how much interest people have.

So far, I managed to build a UI dashboard using Redash connected to my dataset backend:

Next step is to expose to public and implement rate limiting. Will update as I make progress.



Strategy to optimize gain from playing 5 slot machines[1]


  • It’s a Reinforcement Learning framework to solve explore/exploit problems.
  • Eg. What strategy to utilize to optimize returns when playing N slot machines each with unknown distributions of expected returns?

Example Solutions:

  1. epsilon = 0.1 : use 10% of coins for explore, rest to exploit the best known slot machine
  2. UCB method

Further discussions:

  • How is RL different from regular ML?
  • I wonder if Facebook utilizes this in any of their ML systems.
  • Netflix uses this for their recommendation systems[2]

[1] https://www.youtube.com/watch?v=bkw6hWvh_3k

[2] https://www.youtube.com/watch?v=kY-BCNHd_dM



Jerry Hu's brain dump

Jerry Hu's brain dump

previously @ Earnin/WhatsApp/Facebook. I’m interested in Data, Backend, Product, Venture, Web3.