Max Andrews
@thatsjustlikeyouropinionman.com
Father. Pragmatist. Generalist. Builder of things both physical and digital.
im not sure how to do arbitrary relationship extraction from random unstructured text without using any kind of machine learning approach. Feel free to submit a PR if you’ve got something you’d like to contribute.
November 17, 2025 at 7:25 PM
im not sure how to do arbitrary relationship extraction from random unstructured text without using any kind of machine learning approach. Feel free to submit a PR if you’ve got something you’d like to contribute.
D3 force github.com/maxandrews/E...
GitHub - maxandrews/Epstein-doc-explorer: a graph explorer of the Epstein emails
a graph explorer of the Epstein emails. Contribute to maxandrews/Epstein-doc-explorer development by creating an account on GitHub.
github.com
November 17, 2025 at 4:19 PM
D3 force github.com/maxandrews/E...
it actually has about 95% of the value which is fine if speed matters, which in this case it does. I do plan to add a wiki-style suggest edits feature so that refining the data can be crowdsourced.
November 17, 2025 at 4:00 PM
it actually has about 95% of the value which is fine if speed matters, which in this case it does. I do plan to add a wiki-style suggest edits feature so that refining the data can be crowdsourced.
This is also why each relationship is sourced to the underlying document in the UI with subject highlighting — so that you can verify the accuracy at the source.
November 17, 2025 at 3:58 PM
This is also why each relationship is sourced to the underlying document in the UI with subject highlighting — so that you can verify the accuracy at the source.
Sorry that I did not personally extract 23,000 relationships by hand. Feel free to do that and I’m happy to port over to your manually curated dataset.
November 17, 2025 at 3:54 PM
Sorry that I did not personally extract 23,000 relationships by hand. Feel free to do that and I’m happy to port over to your manually curated dataset.
ah cool idea! will work on it
November 17, 2025 at 3:47 PM
ah cool idea! will work on it
What is the logic of this strategy? The top performers flee because they have options (I personally did this) and then everyone else goes through the motions waiting to get zapped. Great way to turn your company into an unproductive husk where everyone is miserable.
November 17, 2025 at 3:43 PM
What is the logic of this strategy? The top performers flee because they have options (I personally did this) and then everyone else goes through the motions waiting to get zapped. Great way to turn your company into an unproductive husk where everyone is miserable.
this is the database file github.com/maxandrews/E...
Epstein-doc-explorer/document_analysis.db at main · maxandrews/Epstein-doc-explorer
a graph explorer of the Epstein emails. Contribute to maxandrews/Epstein-doc-explorer development by creating an account on GitHub.
github.com
November 17, 2025 at 3:27 PM
this is the database file github.com/maxandrews/E...
I’ll add a link to the repo soon which contains the sqlite database with all the data: github.com/maxandrews/E...
GitHub - maxandrews/Epstein-doc-explorer: a graph explorer of the Epstein emails
a graph explorer of the Epstein emails. Contribute to maxandrews/Epstein-doc-explorer development by creating an account on GitHub.
github.com
November 17, 2025 at 3:23 PM
I’ll add a link to the repo soon which contains the sqlite database with all the data: github.com/maxandrews/E...
correct! Though there are certain to be omissions; currently it’s up to the model to determine which relationships are worth extracting and which are noise, and while it does a decent job, it’s going to miss some things.
There are ways to improve i.e. doing a second pass, but gets expensive.
There are ways to improve i.e. doing a second pass, but gets expensive.
November 17, 2025 at 2:00 PM
correct! Though there are certain to be omissions; currently it’s up to the model to determine which relationships are worth extracting and which are noise, and while it does a decent job, it’s going to miss some things.
There are ways to improve i.e. doing a second pass, but gets expensive.
There are ways to improve i.e. doing a second pass, but gets expensive.
No problem to add more documents over time, it will just layer in more connections to the graph. The edges (connection lines between entities) have a timestamp if one was evident in the text i.e. email send date, so you can easily “window” the graph to look at a specific period. Deploying that soon.
November 17, 2025 at 1:54 PM
No problem to add more documents over time, it will just layer in more connections to the graph. The edges (connection lines between entities) have a timestamp if one was evident in the text i.e. email send date, so you can easily “window” the graph to look at a specific period. Deploying that soon.
Think you pasted the wrong link, I got ya: github.com/maxandrews/E...
GitHub - maxandrews/Epstein-doc-explorer: a graph explorer of the Epstein emails
a graph explorer of the Epstein emails. Contribute to maxandrews/Epstein-doc-explorer development by creating an account on GitHub.
github.com
November 17, 2025 at 1:47 PM
Think you pasted the wrong link, I got ya: github.com/maxandrews/E...
A node is an entity (person, place, organization) and the connections are the record of how and when they are connected, modeled from what’s evident in the emails. If subject is unknown or redacted, a placeholder is used.
November 17, 2025 at 1:46 PM
A node is an entity (person, place, organization) and the connections are the record of how and when they are connected, modeled from what’s evident in the emails. If subject is unknown or redacted, a placeholder is used.
Yes, LLM to extract the triples. Though I’m not sure *any* method can give you verifiable accuracy because you won’t have a ground truth. Even a manual relationship extraction will also have mistakes and omissions.
I do plan to add a wiki-style crowdsourced audit tool to flag snd correct errors.
I do plan to add a wiki-style crowdsourced audit tool to flag snd correct errors.
November 17, 2025 at 1:36 PM
Yes, LLM to extract the triples. Though I’m not sure *any* method can give you verifiable accuracy because you won’t have a ground truth. Even a manual relationship extraction will also have mistakes and omissions.
I do plan to add a wiki-style crowdsourced audit tool to flag snd correct errors.
I do plan to add a wiki-style crowdsourced audit tool to flag snd correct errors.
It’s not just a trend — it’s a reawakening.
November 17, 2025 at 1:38 AM
It’s not just a trend — it’s a reawakening.
ah ok i followed back should work now
November 16, 2025 at 5:42 PM
ah ok i followed back should work now
Not sure what that would entail in terms of time commitment, but feel free to DM!
November 16, 2025 at 4:40 PM
Not sure what that would entail in terms of time commitment, but feel free to DM!
This is a massive legal risk to the university as well, and most have strict zero-tolerance policies about any relationships between people in supervisory situations for this reason. In this case, he ADMITS DIRECTLY to retaliating on account of her not accepting his advances by ending mentorship.
November 16, 2025 at 5:33 AM
This is a massive legal risk to the university as well, and most have strict zero-tolerance policies about any relationships between people in supervisory situations for this reason. In this case, he ADMITS DIRECTLY to retaliating on account of her not accepting his advances by ending mentorship.
I hope she sues the FUCK out of the university. This is an insta-settle for 7-8 figures. It proves direct retaliation for not accepting a sexual advance.
November 16, 2025 at 5:28 AM
I hope she sues the FUCK out of the university. This is an insta-settle for 7-8 figures. It proves direct retaliation for not accepting a sexual advance.
Gross, weak, fragile old men.
November 16, 2025 at 5:25 AM
Gross, weak, fragile old men.