No reviews yetBe the first to leave a review for DEHY
Maker
📌
I'm Zayd. I built DEHY.
It started because I wanted one place to see what's actually happening in the market the moment it's filed—not 20 minutes later on a terminal that costs $24k a year. So DEHY watches SEC EDGAR and turns every filing—insider trades, 13F and 13D positions, 8-Ks, IPOs, and earnings—into clean, structured data within about 60 seconds, then puts a 0–100 conviction score on every insider trade.
The part I'm most into: it's MCP-native. You can query the whole thing straight from Claude, Cursor, or ChatGPT. Ask, "What did insiders buy this week?" or "Show me NVDA's institutional flow," and it answers from live data.
I also publish my own research honestly, including the ideas that don't work, at desk.dehy.io. Most trading signals are garbage, and I'd rather say so than oversell.
It's in beta and free while I build. Code **DEHYBETA** gets you in—no card. I'd genuinely love to hear what's missing for your workflow. Reply here or email me.
Report
How are you parsing unstructured text from 8-Ks to surface the signal that matters, and is that logic customizable for my own watchlist filters?
Report
Maker
@rzgarybrp Yeah, this is the part I care most about, so the honest answer is: I lean on structure that's already there instead of throwing an LLM at raw text.
8-Ks aren't really unstructured. They're filed against the SEC's Item taxonomy, Item 5.02 for leadership changes, 2.02 for results, 1.01/1.02 for material agreements, and so on.
So I split each filing on its Items first, then run targeted deterministic extraction per Item type. A 5.02 gets parsed into a structured executive-change row: who, what role, in or out, effective when. "The signal that matters" is mostly materiality ranking. a leadership change or a real agreement over the Reg-FD boilerplate — plus who filed and how it clusters with other activity. An LLM only phrases the summary. It never decides the signal, so it can't hallucinate a number.
On customization: today you can filter by ticker, filing type, and the conviction score. The flexible path is MCP: you query it straight from Claude or Cursor with whatever criteria you want ("8-Ks with leadership changes in my watchlist names this week"), which is more powerful than any fixed filter UI
Report
The sub-60-second ingest on Form 4s is genuinely impressive, filings were showing up faster than my usual Bloomberg workflow. Eager to see how it handles the messier 8-K parsing during earnings season.
How are you parsing unstructured text from 8-Ks to surface the signal that matters, and is that logic customizable for my own watchlist filters?
@rzgarybrp Yeah, this is the part I care most about, so the honest answer is: I lean on structure that's already there instead of throwing an LLM at raw text.
8-Ks aren't really unstructured. They're filed against the SEC's Item taxonomy, Item 5.02 for leadership changes, 2.02 for results, 1.01/1.02 for material agreements, and so on.
So I split each filing on its Items first, then run targeted deterministic extraction per Item type. A 5.02 gets parsed into a structured executive-change row: who, what role, in or out, effective when. "The signal that matters" is mostly materiality ranking. a leadership change or a real agreement over the Reg-FD boilerplate — plus who filed and how it clusters with other activity. An LLM only phrases the summary. It never decides the signal, so it can't hallucinate a number.
On customization: today you can filter by ticker, filing type, and the conviction score. The flexible path is MCP: you query it straight from Claude or Cursor with whatever criteria you want ("8-Ks with leadership changes in my watchlist names this week"), which is more powerful than any fixed filter UI
The sub-60-second ingest on Form 4s is genuinely impressive, filings were showing up faster than my usual Bloomberg workflow. Eager to see how it handles the messier 8-K parsing during earnings season.