Shebby

In build mode

[Shebby @shebby_r]

Mentally thinking that baselining at a singular metric of Revenue Per Token (I'm sure there is more granularity needed here) is where a new business needs to optimise for. Especially as more of the automation will be handled by agent factories!

[Shebby @shebby_r]

@nateberkopec Already building this out for my org.. tethered to a terminal is not good DX!

[Shebby @shebby_r]

Opus 5 constantly misleads by saying it's doing something when I need to ask it to follow through with it. Consistently too.. which was making me question the harness this was running in. Turns out raw sessions are equally bad.

[Shebby @shebby_r]

@AntonKreil One of my favourites Anton!

[Shebby @shebby_r]

Now on to trying to visualise the universe I have access to as the backtest was running limited positions.. This gives me a good understanding of what’s happening under the hood.

[Shebby @shebby_r]

@jobergum Keen to see how this discussion unfolds, following :)

[Shebby @shebby_r]

Some of the earlier work around setting up a DB with prices etc is now paying off. I’m now able to visualise what price action was doing when a trade was placed and really start to think about what my initial filter criteria should be from a TA viewpoint.

[Shebby @shebby_r]

Added a vol sizer to the trading project and now i have drawdowns under control at the cost of performance of course. However fairly pleased that my backtesting pipeline is relatively stable!

[Shebby @shebby_r]

@trikcode php

[Shebby @shebby_r]

@martin_casado I'd love an invite too. Currently in the thick of some infra related work so super relevant.

[Shebby @shebby_r]

As i dive deeper into the backtest data, i wanted to see what the market was doing on a given day. Built a bit of this today.

[Shebby @shebby_r]

Improved backtest performance a wee bit, sorted out the data issues from yesterday. Added a volume filter and some more logging on the streamlit app.

[Shebby @shebby_r]

Spent today working up a Streamlit app that helps me dissect my backtest logs.. Considering i still do not have a baseline for if the pipeline is good, the only way to do this was a viz check. And benefits already. Time to figure out why SPY data is not being pulled through :D

[Shebby @shebby_r]

Recently made the switch from Poetry to UV and really enjoying the DX!

[Shebby @shebby_r]

Spent the day yesterday getting vLLM deployed using ECS. There is a GPU enabled task which seems to work ?! Now to see how this “scales” for heavier workloads. I probably will need to implement a backoff mechanism since it takes a few minutes for the container to be spun up!

[Shebby @shebby_r]

Coming back to the project after the holidays. This looks too good to be true and it probably is. Currently getting into the weeds of the trade ledger and seeing what the crack is. I think i’ve also reached the limit of o1 understanding what it is that i’m asking for it to do.

[Shebby @shebby_r]

And the trading project is growing well. I’m so glad i decided to bake in some unit tests from the start. Got out of many a hole with o1!

Now debating on parquet, csv or sqlite before I do some serious e2e test runs.

[Shebby @shebby_r]

Re-started the finance project. Integrated one single macro datapoint - ISM Manu PMI to a long/short setup. Python, UV, o1 and RepoPrompt. Fun!

[Shebby @shebby_r]

@OptionBit_Trade Useful to see this Steve and to compare to mine from the weekend. Broadly aligned! :)

[Shebby @shebby_r]

Spent the last week or so trying to codify my brain from while trading options over a 3-6 month period. Right now focused on working out the best risk:reward setup given a long and short spread spread trade. 1/8th the work done!