Iβve completely switched my default mode of operation this past year from saving as much as possible to just simply make more.
It happened naturally for various reasons; the most prominent being running the HFT startup.
Quite the difference on how I operate now. Iβd say itβs a
Sasha
930 posts
Performance Engineer. Tokyo based π―π΅
Tokyo
Joined September 2022
- Looks like I'll be moving to Tokyo. I got offered a solid position as a perf & platform eng for Rust trading eng π
- Anyways, I'm happy to announce I've gone full time on the HFT project
- Back on the HFT: - we started to lose money on one of the newer exchanges; figure out why - improve observability; prometheus, grafana, alerting blah blah - scope / benchmark using QUIC over UDP for firehosing market data from one region to another - play around with perf &
- Went from working 7 days a week on HFT and mmβing from a cushy villa to staying in a caravan in the Arctic circle because itβs too cold to camp, oh how times have changed
- Since HFT paused, doing something Iβve always wanted to do. Taking a Yamaha Tenere 700 to the Arctic Circle π Currently in Denmark. Nice place.
- Weβre up ~400% over the last month or so. Weβre slowly increasing capital as we gain confidence. Profits still increasing linearly and we have reason to believe we have some room to grow. The params weβve trained are optimised to reduced large drawdowns at the cost of max
- HFT update: whilst weβre in a good position, taking a break to do some traveling and life stuff. Likely going to spend this time also building some indie products in public for fun. And then back to HFT later this year.
- Today in HFT land: Since we want to A/B test subacounts we need to refactor the architecture pretty significantly as we want to run on the same machine. We have an bucketing algorithm that distributes groups of instruments onto a set, C CPU cores. Now we need to partition C
- HFT today: - We've reached a bit of an impasse. Our main exchange we have been using for testing executions isn't quite as profitable as we expected. And we didn't achieve statistical significance when A/B testing the new setup / param set - However, we believe we will make more
- Today in HFT world: - start writing execution paths of the new exchange weβre implementing - scope how weβre going to hyper-parametrise the params that configure a position reducer - more data analysis for the above Other stuff: - started building out an orderbook in go as a
- First solana hft / mev update for today: - Started building out the backtesting framework - Finished repo pattern to collect solana block data - ~3 days worth of public data collected & stored (free) - Added WS impl to run as a background process to collect block events and fill




