Show HN: I made a free tool that analyzes SEC filings and posts detailed reports
signalbloom.ai(* within a few minutes of SEC filing)
Currently does it for 1000+ US companies and specifically earnings related filings. By US companies, I mean the ones that are obliged to file SEC filings.
This was the result of almost a year long effort and hundreds of prototypes :)
It currently auto-publishes for 1000 ish US companies by market cap, relies on 8-K filing as a trigger.
e.g. https://www.signalbloom.ai/news/NVDA will take you to NVDA earnings
Would be grateful to get some feedback. Especially if you follow a company, check its reports out. Thank you!
Some examples: https://www.signalbloom.ai/news/AAPL/apple-q1-eps-beats-desp...
https://www.signalbloom.ai/news/NVDA/nvidia-revenue-soars-ma...
https://www.signalbloom.ai/news/JPM/jpm-beats-estimates-on-c... (JPM earnings from Friday)
Hallucination note: https://www.signalbloom.ai/hallucination-benchmark
I was thinking of building exactly this type of thing here. This is great and I love the simple/minimalistic styled pages.
Things I was also contemplating to implement you can freely steal:
1. Giving a list of potential interesting investments - these could also be behind paywall.
2. Summarize based on sectors to see where there's downturns coming - again maybe paywall this
3. Connect with news-sites or twitter etc.. to see what is happening and what is the sentiment. This is a lot harder as you'll need to either scrape or pay to get that info.
Best of luck
Really nice of you to encourage, thank you! I've thought about some of those.
The first one is an interesting idea, but offering any real financial advice is regulated, which is a tread carefully territory
Second one is achievable and a good one at that. In fact, there is a lot of canonical data out there that when used carefully can provide a very nuanced picture of the current state of economy.
Third one, currently I have just a twitter account where the earnings reports are auto posted the second they are ready (that is, until I run out of free api limits). Their API cost for at the higher end makes it only viable if I am doing it full time and there was a clear edge by analyzing the data. My gut feeling is, most of the work will go into filtering out low signal posts idk
I'd suggest moving to either bluesky or rss. In fact, selling rss access (very fast responses) and delaying the public (twitter/Bluesky) releases by a few hours/days may be a way to monetize successfully.
Unpopular opinion here... If you tread carefully you'll most likely not succeed. I am not American and I know you guys like to sue eachother for putting cats in microwaves and stuff so maybe this is not great opinion to have in America at the current moment.
I would go for it and put a disclaimer, or I would just incorporate in a country where there's no issues with these things.
all this is hard of course to provide good value, but worthwhile.
Twitters' cost is insane right now, I had quite a few ideas for twitter integrations but they would easily cost thousands per month just to access their API.
I looked into https://github.com/d60/twikit - might not be suitable but you can definitely play around with it. Just don't use your official account as I got shadow banned using it unfortunately.
>Unpopular opinion here... If you tread carefully you'll most likely not succeed. I am not American and I know you guys like to sue eachother for putting cats in microwaves and stuff so maybe this is not great opinion to have in America at the current moment.
It's not actually about personal suits, financial advice may fall into financial services regulations and that's regulated by the government of the US, EU, etc. so wherever the company is registered may have issues in addition to the potential of the regulations applying to the individual and not shielded by the company.
Was figuring out how to do the same thing, but for an Indian context, and specific to one's portfolio, by perhaps reading emails. Was still ideating it, cause I am working on another thing at the moment.
I'm interested in this topic too, but am approaching it from an auditing rather than from an investing angle.
I read lots of quarterly report summaries. I like what you're doing here, but you need to be aware that AI summaries are already built into many existing platforms. So don't expect to be able to monetize this as it stands.
And way before the current wave of LLMs, I did some consulting for a company that provided these kind of summaries over a decade ago, and I’m sure they weren’t that early themselves
Yes, but:
1. These summaries were terrible. 2. The economics of these summaries was to get into Google News, spam newsfeeds, and use this to drive traffic to your paid product. This business model stopped working when Google News stopped accepting new sources for their Finance feed...because it was full of spam (I worked for a company that tried to start in this area with human writers, didn't work because they couldn't get onto Google).
Simply Wall St was one company that made tens of millions doing this, there were many others...but none were doing anything that is related to what OP is doing.
Every time someone will tell you something can't be done...this very clearly can work, it is not easy but ad revenue for financial services is still completely ludicrous and most services doing this are complete garbage (the news articles above the OP is producing are far from it, I worked in equity research, this is better than most journalism in the space...which is now non-existent...and probably at the level of a junior analyst).
When I was graduating college before the Great Recession, several of my recruiter pings were for companies wanting to parse financial documents.
If they were reaching out to not particularly special undergraduates twenty years ago, I'm guessing it goes back at least thirty.
Love it! This is a very useful tool to provide me with introductory (but granular enough) information about new investment ideas that I have.
Just two cents based on my usual workflow --
1. I like to compare companies within the same industry, so it would be good if i can ask follow up questions after reading a report like "How does company Y compare with company X on XX metrics?"
2. I understand that the reports are generated by a model; if that is true, maybe you could cross-reference the fillings with their corresponding earnings call transcripts (should be available for free on many investment websites) and highlight what is being discussed/asked in the calls as well
Again, thank you for the great work. Already pinned your site to my browser's investment workspace XD
Thanks, both are really good suggestions!
Well done site. One important nit pick: never use charts that don’t start from 0 on the y axis. I was looking at a stock that had a yoy growth rate reduction of 6% (from 39 to 33 for each respective yoy period), and the chart showed an aggressive down to the right trend line because the y axis started at 33% instead of 0%.
Charts like that show more detail sure, but everyone freaks out in reaction to them. Always zero out your graphs.
> Always zero out your graphs
Great way to end up with useless graphs where you have a tiny line at the top that's been compressed to the point where you can't see any changes...
The "rule" about ensuring axes are zeroed is for bar charts not line graphs.
Charts like this are standard practice in finance. You're interested in the relative movement, not absolute numbers, so a chart that starts from zero is practically useless since the change you're trying to see is too small.
I disagree with this. Zero is an arbitrary and often useless intersection. A stock worth $300 is going to show meaningless movements at that scale.
This is correct. GP’s rule should be adhered to for bar charts though. Absolute values there are expected. For line charts continuous (ish) over time, people are usually more interested in relative change in a time period so ok if y axis starts closer to the minimum value of the series in that range.
Yeah, I would say the y-axis range on charts should be set at "3-sigma likelihood of observation" thresholds. Not everything that's charted can be framed as sampling from a distribution, but the principle of manually setting chart ranges would nonetheless still apply.
For instance, if we're charting someone's body temperature, we would likely fix our y-axis to 80-110.
That’s the exact point?
No, it isn't. People are interested in whether a stock has volatility, and whether it has moved x%. If a $10 stock loses $1 it should show a roughly matching pattern to a $200 stock losing $20. Intersecting at zero will show a very large movement for the $10 stock and a very small movement in the $200 stock, despite the effect being the same to stockholders.
> Intersecting at zero will show a very large movement for the $10 stock and a very small movement in the $200 stock...
How so? Surely both will show a 10% movement?
Zero is not arbitrary when it comes to the stock price.
It very much is, because stocks don't start at zero and your entry point will also not be zero. You can be deep in the red long before the stock gets anywhere close to zero. Even when a stock hits zero it doesn't always mean something useful, look at the Hertz bankruptcy for a good example. As far as stocks are concerned, zero is arbitrary and pretty universally a useless reference point.
Adding to this, stock prices should be plotted on a log scale since it is log returns that are roughly normal, and then 0 really makes no sense.
One child comments makes a better suggestion. Display only a change in % since period start. Who cares about the absolute price. You only care about percentage change.
And yes, all charts would automatically start from 0 as a side effect.
Change could go to negative though, so even though 0% is included in the chart, the true possible bottom value of the chart is -100%.
I wish you also offered company summaries, not just news.
Form 10-K is usually what you'd need to parse for those, maybe with recent stats taken from the 10-Q.
It'd be really great if one could get a report on a company in an industry they're unfamiliar with, with statistics like profit margin available at a glance, and a good description of how they actually operate, what is important to them, what areas of the business actually bring in the money (this is often surprising!) etc.
That's an interesting idea, thinking about it.
The plan is to address all SEC forms anyways so this will be a natural extension
It seems to be the season for scraping SEC filing archives and running them through AI bots to extract information. I made this:
https://boards.industrial-linguistics.com/
It only looks at information about directors on company boards, and unlike your much better project, I don't have any clear idea how to commercialise it.
It really seems like it is, I made this YouTube channel to do it as well
https://youtube.com/@corporatedecoder
I love it ! Thanks for sharing here !
I have to build a project for data analysis with ML (and optionally AI) for a course and certification I am passing, this is a great inspiration. I will probably do something simpler on European stocks.
You say it was a year long effort, was it full time ? Do you have an estimation of how many hours you spent on it ? Did you do it alone ?
I lost track of the hours lol. Technically it is a side project but I have been putting all my spare time into it. I had some help too.
For European stocks, is there a common EU regulatory body where all filings are submitted? That would be a good start
would be interested to follow along! Where would you source the filings for european stocks?
It’s nice, I like it.
PS if the next question is whether I’d pay for it, the answer is likely no ;)
A suggestion: I forget the HTML tag off the top of my head, but there is something you can do that turns off the keyboard autocorrect. Setting the search box input type to something specific or something like that… stock tickers aren’t dictionary words :)
Thank you! I love this about HN (well, I've been a lurker mostly for 18 years) that people point to things you haven't even thought of.
Looks great! For a long time I wished there was a historical analysis that would show quality of the management. Meaning, what initiatives did the management bet on in the past years and how well did it pan out? If a bet failed, was the management honest about it?
Interesting, good job.
Minor suggestion: I would make the home page feed show popular stocks (so that it is more interesting to new visitors and also engaging to regular visitors). You could just pick 100 stock symbols or combine that with something like what you are seeing people search for (I would manually check the popular search ones - some may not make sense to add to the home page feed).
Seriously fantastic tool, I'm going to use the hell out of this, I appreciate you creating this so much! Let me know if there's a way to donate.
Much appreciated. The only donation I can think of is if you help spread the word. I learned a tough lesson this week that it is extraordinarily hard to market a product lol
P.S. If it wasn't for @dang putting it in the second-chance-pool, this post would have never been seen by more than 2 people.
Great work you’ve done here. I worked on a similar project for about a year but ended up dropping it because it wasn’t fun for me. Mine failed — I hope yours succeeds. Best of luck!
Any pointers on creating similar tools? I had an idea for an AI-backed analyses tool (not for stocks) but not quite sure how to go about it. Even if it's just a pointer to some page that does an ELI5 I'd really appreciate it.
Depending on how much time you have, the first thing I would recommend is to watch Karpathy's 'deep dive into LLMs' video (https://www.youtube.com/watch?v=7xTGNNLPyMI). It is 3.5 hours but it will give you so much instinctive understanding of how these things work and will likely save you a lot of time.
Purely from a technical standpoint, things aren't that difficult really. I'd recommend get yourself an API account with all major providers (because each frontier LLM evidently has their own strengths and weaknesses) and start experimenting with your data. The feeling you'll get when you first get an API response will keep you going! :)
Having said that, a caveat is, LLMs do hallucinate from time to time. If your data needs accuracy, definitely don't do the "throw everything at LLMs and let them do the work" but I'm sure you'll figure it out from trial and error. Good luck!
Will check it out. Thanks so much!
I'm pretty familiar with how LLM's work, it was more of "how do I create sites/apps" with LLM/AI backends. Will check out the youtube. Thanks again.
> I'd recommend get yourself an API account with all major providers (because each frontier LLM evidently has their own strengths and weaknesses) and start experimenting with your data.
Wouldn't it be better to use something like openrouter.ai which gives access to lots of different endpoints? In my experiments that is what I am using, wondering if there are good reasons to go to direct API usage.
Yeah Openrouter is generally better. Only exception is when some API is launched but not yet on openrouter, rare though
I love it. I think there is room for this product even though there are other options. This is clean and probably over time you can introduce more features as you get the feedback. Eventually you will be able to monetize but for now listen to your audience.
One thing that I have hypothesized about, is can material embedding vectors be extracted which are then used to predict prices?
In theory, this is what value investors are doing to some extent. However, embedding vectors often to do match intuition very well.
Pretty cool, reminds me of content from SeekingAlpha.
I'm trying to do something unrelated with public companies, where did you get a list of all companies filing with the SEC? Or was it built based on whoever filed 8-Ks? A google search leads me to this list [0] but it's missing a few companies.
[0] https://www.sec.gov/files/rules/other/4-460list.htm
The EDGAR website has a symbology of CIKs where all companies, people and events are listed.
There is an API, and daily historical snapshots available.
Fantastic job, are the charts generated by the LLM itself or you have some kind of postprocess that inserts them? I'm working on a similar problem space and charts are difficult. Thanks.
Very cool! I'm curious how much of this was vibe coded?
Really clean site, I've also been working with similar stuff. If you need inspiration, check out https://findl.top
sorry i dont understand. isn't this information already available on the SEC website? what does your product do
The earnings filings are available on SEC. Where the product makes a difference is by analyzing interesting patterns and contextualizing it with different means
e.g. this NVDA earnings report: https://www.signalbloom.ai/news/NVDA/nvidia-revenue-soars-ma...
beyond the headline revenue/EPS miss/beat which is trivial, it correctly identifies a trend of margin compression and charts it. It also identifies the gaming revenue divergence from the datacenter revenues and so on. You can check out pretty much any report and odds are, you'll find things that weren't called out in the earnings release
I am a portfolio manager of a small/mid-cap equity fund. This is definitely cool and clearly took a lot of time and thought!
If this is a side project, that's amazing and I love it. If you did want to take it further though, I would though encourage you to spend some time interviewing potential users and thinking through how you are creating value.
As a fundamentals investor, to win in stockpicking (especially against quant traders), you need to get really deep into the details. As mentioned there are lots of tools that summarize earnings calls, but more importantly, I don't think summarizing earnings calls or releases are super useful for investors.
There are basically two ways to build useful software for investors. One way is to create summaries / off-the-shelf analyses that tell investors if a company is good or bad. Right now as-is, that is sort of where this product is. The problem is, an earnings call summary is insufficient. Yes, AAPL revenue may be up 4.0%, but the fact that you figured that out in 15 seconds with an LLM is not a differentiator since everyone else can figure it out in 45 seconds by just doing mental math. And, what did _you_ expect it to be? Very different trade answer if you thought it should be down 10% or up 40% and it in reality hit 4.0%.
I think generally this is also tough because the people really good about building recommenders (black box or otherwise) just work at hedge funds and make a ton of money for themselves / their clients. It is almost always going to be more profitable to do this, in no small part because once you reveal your secret sauce to the market, it gets priced in.
There's also the issue of user personas. There is no one generic investment strategy. A "good" investment is very different for a long-only vs. a long-short vs. a macro fund vs. an activist fund (and yes these are not mutually exclusive at all).
So, I think in order for this to become truly useful you need to make it more of an interactive tool to allow people to run their own analysis (vs having a "one size fits all" analysis). A few ways you could do this:
1) investments are done behind an investment thesis. Can you TRACK theses over time? For example, a merger arb thesis might be that a big M&A deal is on track (or not on track) to close. Can you track my THESIS, which almost becomes a detailed LLM prompt. E.g., "I think that Company X's acquisition of Company Y will close at the current offered price of $Z by date A, and key milestones are (1) regulatory approval on or around date B, (2) stock price converging toward the offered price of $Z as date A approaches, (3) positive sentiment on earnings calls about the acquisition by both Company X and Company Y"; let me know if earnings releases or calls or 8-Ks should make me update my thesis ("8-K came out that said that the target date of A has slipped by 90 days"). Or, "Company X said they bought Company Y and at the time expected revenues of $Z in three years; what compound quarterly growth rate does this imply" and then track that every quarter to see if they are on track or not
2) Can you help me track a specific question through each quarter—almost like automated BI / reported? Bloomberg/CapIQ/etc. will do this for GAAP metrics like revenue or COGS but often companies will report bespoke metrics. Can you help me track WeWork's community-adjusted EBITDA (lol) every quarter? Can you help me track business unit non-GAAP EBITDA with an addback for taxes every quarter? These are things that won't show up in your GAAP 3-statement model but for _individual_ companies may be super important
3) similarly can you help me run comparisons across companies? E.g., can I compare Vornado's total revenue (almost all office) to Bridge's office segment's revenue to SL Green's total revenue (also almost all office) and every quarter see how their relative revenue (market share) is changing? What about their occupancy and rent levels?
4) generic RAG/natural language search (when did SmartRent talk about their acquisition of Planon), although as noted, many competitors are doing this
All of the above are things that analysts do in real life and basically just track manually in Excel.
Again don't mean to be discouraging—I posted this in part because I think there is a ton of value here to (a) having a clean pipeline and (b) having a basic working knowledge of applying LLMs in a useful way. If helpful I'm happy to discuss more.
Kindly let me know how to get in touch with you! Alternatively, feel free to message me, contact info on profile
Seems like 1-3 above could be answered/tracked with some plumbing pipeline elbow grease leveraging tools like open source edgartools[1] (just a happy user) or sec-api[2] paid proprietary service.
4 would be a heavier lift because it goes beyond the basic skill set of an average data/junior analyst.
[1] https://github.com/dgunning/edgartools
[2] https://sec-api.io/
Superb comment! Appreciate the insights and the time you took to write it. If possible, I'd like to get in touch with you to talk more about this. Couldn't find your contact info on your profile
This is cool. If you don’t mind me asking, are you using the EDGAR api or something else?
Thanks! No we (had some help) built the full pipeline from the ground up using EDGAR archives. Cleaning up SEC data was a nightmare but we eventually got there
Fascinating. It seemed like the EDGAR API looked too good to be true. Sounds like it was.
From what I gather, there is an "official" EDGAR API that is not operated by SEC but by a company they authorized, costs tens of thousands per year and by some accounts, is not that great.
Then you have a bunch of third party providers that offer APIs that return cleaned up SEC filings and charge like few thousand dollars a year. The downside is the latency they add.
Yeah, it's a SCP/SFTP service managed by a consulting company:
https://www.sec.gov/search-filings/public-dissemination-serv...
Not sure why, your website is blocked by Office's Cisco Umbrella.
Your search seems broken (maybe because of the HN hug)
Also, rename "News" to "Latest Filings" or something more direct. Generally websites use "news" to refer to news about themselves, which nobody cares about.
This is a really good suggestion (changing "news" to "latest filings"), done, thanks!
Is anyone trying to automate the detection of the rampant insider trading that is, no doubt, going on at present?
It's pretty trivial to find insider trading looking activity as a human, so I doubt it would be particularly difficult to write up an algorithm to do it. Of course you wouldn't come close to catching everything but you'd catch blatant stuff.
The problem is as a private citizen/company (1) you have no idea who placed the trades and (2) even if you did, you can't enforce the law.
And, as seems to be the case with U.S., even the justice department themselves have little interest in enforcing the law, making insider trading the sort of thing that on paper is illegal, but since everybody does it, it's okay.
I have a prototype doing just that! That is on the roadmap :)
Although, it is not as exciting as some make it sound. Most insider trading reports are utterly boring. There are a small percentage that are valuable though
Unusual Whales tracks as much as they can... https://unusualwhales.com/insiders/trades?transaction_codes[...
Finding insider trading is not the difficult part, it's being able to prosecute it that is going to be a challenge.
You might like edgartools by Dwight Gunning
What’s your stack?
The way to make money from this is to auto generate content and use SEO. There is an incredible amount of money in this niche and many players making money doing similar things.
Curious what you've had in mind here regarding auto generated content and using SEO? How is this supposed to make money?
Out of curiosity, how does your tool manage to account in its reports, for entities that have engaged in keeping a separate set of books as covered under FASAB S-56?
If you are unfamiliar this was passed mostly secretly in 2019 during the whole Kavanaugh confirmation, almost without issue crossing party lines. Its made every SEC required statement effectively worthless for the entities involved in any project or work that touches on national security.
There are at least 7 of the largest companies that make up the majority of the DOW/other indexes that may be covered.
Currently it does not track the separate books. Something that needs to be done in future iterations. Thanks!
I've been wanting to build a stock trading bot but my general findings were that stock move direction was largely uncorrelated with news of these sorts. Moves happen, but unfortunately it often happens that really good news can be put out and the stock just crashes.
I swing trade on the side, and tbh news don't matter - manually all I do is have an opinion on an absurd price range, I do a buy order at the low end, it eventually fills, I set the sell order at the high range, and one day it'll fill. If I want to earn while waiting, I do options.
Automate this maybe ? Don't follow the news because you'll never fully understand the implications: firing half the employees could be good news because it relieves cash stress or bad news because it means no more sells of the company's product. Better to at least just look at the post-news trend and position with it?
I think most people end up with a timing issue: they want to predict time of the change of movement but why don't they just predict the price ? Remove time, and you'll never "lose money" - but your time-value will maybe be less efficient than positioning on an ETF and waiting.
Basically, if one could do a trading robot making profit overall, based on public news, it would already be done, and therefore you would be unable to compete. Flash news, it's already been done. You're like a little cell appearing in an ocean of sharks, the evolution of your competition already outpaced you.
I’ll take a look, thanks for the effort!
(Side note: would you mind emailing hn@ycombinator.com so I can explain how and why I edited your text here?)
Done. Thank you!