Every time we have a new model, I run the same simple prompt on all my projects:
"Read this entire project and find everything that’s broken, inconsistent, inefficient, or no longer needed. Fix what you’re sure about and list the rest for me to review."
highly recommend trying
Some important things to understand about market structure:
-fundamentally nothing has changed since 10/10 re: alts and arguably has gotten worse
-as such, liquid funds can’t size as big on alt allocations. This was the main structural bid on “quality alts” in the past. Most people are just trading momentum/trend
-Capital / margin is more spread out because of the growth of cross-marginable stock / RWA perps. Fiat stables haven’t grown their market cap in some time now
-leakage: Hacks, exchange fees, launchpads, grifters. Same as before but recent uptick in hacks due to AI cyber capabilities
-so who has money? (1) Basis/systematic traders who got favorably ADL’d on 10/10 ie not a source of bid on market beta (2) HYPE airdrop recipients/holders, who are mostly losing traders / momentum types (3) hackers who sell coins -> fiat
These aspects all contribute to crypto trading like absolute garbage outside of some short periods where people are forced to chase aggressively trending coins
People have been asking for my AI setup for financial markets so here it is:
Set up Hermes -> Get it set up with yfinance price pipelines -> Hook it up to SEC EDGAR / other news outlets you want, might have to pay -> Can get external / faster APIs from paid data providers for whatever you need
Why: Its important to hook up the AI to real data if not it will hallucinate
Afterward, set up the choice of interface - Terminal / Telegram / Discord / etc. all works
Next, you'll need a structured template. I suggest generating one yourself using Claude Opus 5, and then edit it to your hearts content.
The steps after this are all tinkering - just adjusting the kinks, adding things you like (charts, etc.) Using the template, telling the AI to generate the report in that style when you ask for it, earnings report for your holdings when you ask for it, etc.
having been on multiple desks I cannot emphases the importance of continuing to improve the trade. having personally been on desks trading very similar variants of strategies, its always interesting to see how widely pnl varies from the firms have put in to bettering the trades
New traders get told that they have to have specific 'setups'
So they spend all day back-testing some random price pattern and shove a couple of indicators in there for good measure, and then fire risk into the book as soon as it 'triggers'
In reality the vast majority of your returns as a crypto trader most likely won't be set up-driven but rather regime-driven
Most of the time the market does mostly nothing
And if you try to force your set ups in the wrong regime, then you'll just get bad results (e.g. momentum trading breakouts in a choppy market)
So I think the first useful port of call is at least some broad sense of what type of regime the market is in
This doesn't need to be complicated: you can just look at the chart, use some basic moving averages or even common sense to get an understanding of trending, mean-reverting, momentum, catalyst-based, volatility compression/expansion etc (complexity can be layered in later if required)
Each regime should then have specific playbooks or set ups associated with it
For example if the market is trending then you might adjust to trade shallower pullbacks, lower time frames, faster moving averages, less drastic momentum resets etc.
That same so-called set up would get you absolutely cooked in a choppy market
In other words the super TLDR is that the correct hierarchy is:
1. Regime is the first filter
2. Setup is tied to the regime
3. Execution nuances
As a practical example:
The current regime is technically defined as "soul-crushing dogshit" so the correct setup is to develop a nicotine and video game addiction
Hope this helps
LLM Knowledge Bases
Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So:
Data ingest:
I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them.
IDE:
I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides).
Q&A:
Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale.
Output:
Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base.
Linting:
I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into.
Extra tools:
I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries.
Further explorations:
As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows.
TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.