Most active commenters
  • sand1929(15)
  • bozhark(3)

42 points sand1929 | 41 comments | | HN request time: 1.355s | source | bottom

Hi HN! I’m Sandy, and I’m excited to introduce Parsagon. Parsagon is using AI to automate workflows for government affairs professionals, starting with an AI search for public policy. Here's a demo: https://parsagon.io/explore-search

Current search portals for political monitoring are outdated. (By political monitoring, we mean people/orgs monitoring government announcements, policy updates, etc.; not the government monitoring people.) Each one only covers one region of the world, and they only allow you to search government publications based on simple keyword searches. Government affairs professionals often spend hours combing through noisy search results to find relevant developments. For example, they might search for the words “labor”, “employment”, “workforce”, etc., combing through all US government publications mentioning these keywords to find relevant material. Then they might have to repeat this process on separate platforms for the UK, EU, etc.

Parsagon allows you to search for exactly what you want. Instead of searching for all publications containing the phrase "crop production", you can search for something as precise as "news related to US crop production, including foreign crop production that impacts the US" and get the publications you’re looking for from any region of the world (our demo lets you search the US and UK).

So you might be wondering: why are we launching on Hacker News now? After all, we’re YC W21, and we’ve been around a while. Well for the first 2-3 years of our existence, we were working on an AI developer tool that generates data pipelines, and while we had plenty of people interested in using our AI, our AI couldn’t solve most of the use cases we encountered (at least, it couldn’t solve them well). Everyone’s use cases were quite disparate, and we wasted a lot of time trying to get our product to the point where we could solve all of them (and during that time, we just weren’t concerned with launching on HN).

We had little success until a large non-profit reached out for help getting started with our product. They wanted to scrape a wide variety of government organizations to get announcements and communications that could affect their line of work. This request seemed odd to us, since there are already platforms that track government activity seemingly ready-made for their use case. Why would these mostly non-technical government affairs professionals be trying to learn to use a developer tool?

As we talked with these first customers and learned how outdated current political monitoring tools are, this use case became increasingly exciting to us for a few reasons. First, we felt our AI gave us a significant advantage in this space, allowing us to monitor government websites on a scale that existing tools couldn’t match. While most existing products focused on monitoring a single country, we could build pipelines to get most of what our users wanted from any given country in under a week, and give them better ways to search and aggregate that data at the same time.

Second, two other significant organizations reached out around the same time with similar use cases. We were excited that larger organizations were interested in this use case and were willing/able to pay significant sums for it (one problem we had before was that many companies interested in using Parsagon were small startups with low willingness/ability to pay). And since these organizations were able to introduce us to others in their industry, it quickly became apparent that we should focus on solving this one use case of political monitoring.

And so now we’re here! We have a product we’re excited about, that our current customers are excited about, and we want to share it with HN! We’d really appreciate any feedback, and if you know anyone working in government affairs, we’d appreciate it if you showed this to them!

1. stuartjohnson12 ◴[] No.42319484[source]
Congrats on the launch!

I know it is truly the peak of the spirit of Hacker News to make the first comment a nitpick, but the fact that your name is a play on Pentagon but your logo is a hexagon causes me irrational levels of emotional distress.

replies(3): >>42319572 #>>42319949 #>>42323150 #
2. sand1929 ◴[] No.42319572[source]
LMAO, this is unironically the kind of feedback I love to see from HN
3. mung_daal ◴[] No.42319761[source]
LLMs do not know what a number it is. public affairs and government are bean counters. these are conflicting ideas that any regex script can handle better.
4. ◴[] No.42319825[source]
5. instalabs ◴[] No.42319871[source]
Congrats on the launch - there's a lot of unstructured public data as well (existing in the form of PDFs/Excel/PPT) are you planning to add support for those?

Also would be great to show the source in the list without having to click "source" :)

replies(1): >>42319925 #
6. sand1929 ◴[] No.42319925[source]
Absolutely! We're already doing this with PDFs from legislatures for our tool to track bills, which should be ready soon.

Also, thanks, that's great feedback :)

7. marban ◴[] No.42320225[source]
You're RAGing over a DB of scraped news articles?
replies(2): >>42320309 #>>42320342 #
8. ada1981 ◴[] No.42320259[source]
We were just discussing this TODAY in our weekly AI Playground meetup; that someone will soon make something like this.

I’m curious if private ngos or companies can use it as well? Can we track policy and potential impacts on industry? Etc?

replies(2): >>42320379 #>>42323087 #
9. enisberk ◴[] No.42320280[source]
Congrats on the launch! While search with LLMs is quite popular nowadays, it is still hard to get it right.

As a smoke test, I tried the following queries, and they returned the same result. Good job!

    international relations Turkey
    international relations about the country with the capital city of Ankara
Both return info from this link: https://www.state.gov/secretary-blinkens-call-with-foreign-m...
replies(2): >>42320400 #>>42320431 #
10. sand1929 ◴[] No.42320309[source]
News articles, Requests for Comment and similar government announcements, speeches, parliamentary questions, government reports/statements/consultations--a pretty wide variety of sources
11. zxexz ◴[] No.42320342[source]
I do this with the federal register + the comments and attachments. Probably the most informative govt. website.
replies(1): >>42322187 #
12. sand1929 ◴[] No.42320379[source]
Yeah absolutely! We have private companies using us right now! (We might have an private NGO as a customer--I'm always a little hazy on the exact definition of NGO)

I definitely think we're going to see a lot more people building in this space. There seems to be a growing consensus among the government affairs people I've talked to that it's just a matter of time

13. sdesol ◴[] No.42320400[source]
Well if you ask GPT-4o the following:

Give me keywords to search for based on this sentence "international relations about the country with the capital city of Ankara"

You get the following:

- Turkey international relations - Ankara diplomacy - Turkey foreign policy - Turkey global partnerships - Turkey international politics - Turkey geopolitical strategy - Turkey foreign affairs - Turkey global relations - Turkey NATO relations (if relevant to your topic) - Ankara as a diplomatic hub

So it is not unsurprising that the same link was returned

replies(1): >>42322737 #
14. sand1929 ◴[] No.42320431[source]
Thank you! It's been interesting to watch HN playing around with it. This community definitely phrases its search queries differently from how many government affairs professionals would (especially to try smoke tests like yours), so I'm glad it's holding up :)
15. _1 ◴[] No.42321052[source]
Is there a custom search engine underneath this? My company's CIO has been asking about an LLM-driven search engine for the entire company.
replies(2): >>42321386 #>>42322192 #
16. sand1929 ◴[] No.42321386[source]
If you're talking about search-as-a-service solutions, we didn't end up using any of them. There were certain optimizations we needed to make to our information retrieval algorithms that were pretty specific to political monitoring, and none of the out-of-the-box solutions seemed customizable enough for us.
17. kianN ◴[] No.42321434[source]
Congratulations on the launch! A tangential use case to your current focus could also be smaller financial firms focused on fundamental analysis. I've done some work in the past setting up news datastreams for these sort of firms. Something like this that structures public government data could be really interesting for them.
replies(1): >>42321558 #
18. 1123581321 ◴[] No.42321488[source]
This is neat; congratulations on launching. Curious what you do to assure users that you’re searching every corner they’d be checking manually, and that there haven’t been any regressions to the source set. This came up at work recently.
replies(1): >>42321726 #
19. sand1929 ◴[] No.42321558[source]
Oh interesting, I'll be sure to look into that! Definitely an intriguing use case
20. christoff12 ◴[] No.42321629[source]
Love the PMF journey -- good luck!
21. ned ◴[] No.42321651[source]
I see that you are quoting in extenso News articles (I read one from the DailyMail). How did you secure the copyright issues of reproducing full articles ?
replies(1): >>42321889 #
22. sand1929 ◴[] No.42321726[source]
Thanks! I think the best thing we do to assure users that we check everything they'd check is that we add requested data sources pretty fast. E.g., one customer wanted to track announcements from the various state/regional commissions on higher education, and we added those within the hour.

For political monitoring in particular, I think this is pretty important. Different organizations in different industries will inevitably have very specific/obscure sources they want to track, so rather than promising that we already have everything they'd be checking, we just assure them that we can add anything missing really quickly.

23. abe94 ◴[] No.42321861[source]
Love the story about your journey sandy - its been great watching you stick through the pivots!

Could your software be used to create something like a specialized feed? For example like the newscatcher feed product (if you are familiar with them)

replies(1): >>42321953 #
24. sand1929 ◴[] No.42321889[source]
Oh whoops, that seems like a bug. We typically show press releases verbatim, but our intention is to just provide the headline, link, and extracted facts for sources like the Daily Mail. Better see what's going on there; thanks for the catch!
25. sand1929 ◴[] No.42321953[source]
Thank you! And yeah, getting specialized feeds is one of the main things our current customers do!

Also, hope all is well at Laudspeaker :)

26. bozhark ◴[] No.42322179[source]
Our use case is similar to your "first" customer

at first we got zero results. tuning specific wording from "requirements" to "policies" got the intended results

thank you for this, excited to try it further

replies(1): >>42322362 #
27. bozhark ◴[] No.42322187{3}[source]
got a github to share it?
28. bozhark ◴[] No.42322192[source]
same, anything suggested?
29. sand1929 ◴[] No.42322362[source]
Glad to hear it!

If you have any questions or need anything, please reach out at https://parsagon.io/contact ! Working with non-profits has been some of the most fun work we've done, so we'd love to do what we can to make the platform work for you!

30. vm ◴[] No.42322382[source]
Seeing your launch hn made me happy. I tested one of your products in 2021. It's cool to see you and Parsagon working through iterations and finding customer traction. Mad props!
replies(1): >>42322437 #
31. sand1929 ◴[] No.42322437[source]
Aww, thank you! I'm impressed you even remember us from 2021 haha
32. nisten ◴[] No.42322711[source]
It still boggles my mind what mental & legal gymnastics are passed through to get 3 years of funding for what looks like someone last night put together a RAG of a cbp data dump that you get for free from their site.

Not saying it's not useful, just fascinated at what excuses have gone in endless meetings for 3 years to do whats basically an afternoons worth of work.

Yes I mean that literally. You start off an aws rag app or a huggingface container, you download the CSVs fom cbp public portal, and you add a sh*tty backgroundcolor and a border glow that doesn't go all the way around a button. 10-15 more years of this and maybe we'll get the next dropbox again who knows.

replies(1): >>42323617 #
33. enisberk ◴[] No.42322737{3}[source]
This is a demo from a small startup dedicated to enhancing government transparency, which I greatly appreciate. As a result, my expectations are aligned with this goal, and I refer to this as a smoke test.

Achieving accuracy with RAG and LLMs is a challenging task that requires balancing precision and recall. For instance, when you type "Ankara" into GPT-4o, it provides information about Turkey. However, searching "Ankara" in their product does not yield articles related to Turkey.

replies(1): >>42323172 #
34. KaiserPro ◴[] No.42322919[source]
Perhaps I am not using it right, I wanted to see if I could get a list of sources for "policy roadblocks to timely court justice" for the last 7 days in the UK.

I get some interesting articles, but only really one was relevant

even then it was only tangential.

Its kinda useful as a news cutting service, but its really not precise enough to replace a decent PR summary report.

for housing policy It either over indexes on hansard (good source, but not for policy) or the daily mail (would have been a good source for conservative policy, but not labour.)

What am I doing wrong?

replies(1): >>42323183 #
35. jasonsteving ◴[] No.42323067[source]
It's interesting to hear about the twists and turns that got you here, thanks for sharing.

I happen to have built something in the same vein in a few months earlier this summer for the Gemini API developer competition.

For anyone interested, check out Gov Notes (https://app.gov-notes.com)

The site lets you dig into US house committee hearing videos posted to YouTube using AI search and chat. Even have some fairly powerful grounding in the chat - you can ask followups like "when was that discussed in the video?" and the chat will reply with a clickable link to a specific timestamp in the video where your current topic was discussed.

36. jasonsteving ◴[] No.42323087[source]
Check out https://app.gov-notes.com/ - dig into US congressional committee hearing videos (for free)
37. layer8 ◴[] No.42323150[source]
Think of it as a polygon with about 4 km circumference: https://en.wiktionary.org/wiki/parsa#Noun
38. sdesol ◴[] No.42323172{4}[source]
> Achieving accuracy with RAG and LLMs is a challenging task that requires balancing precision and recall

The challenge is domain knowledge and not tech in my opinion. There are dozens if not hundreds of companies providing RAG and LLM, but the challenge is, like you pointed out, what should you do if you encounter something like "Ankara".

For BestBuy, this might not mean much, unless there is a BestBuy in Turkey. For a government related site, cities and geography is important, so trying to extract additional meaning from Ankara is probably important.

39. sand1929 ◴[] No.42323183[source]
So we actually have 2 search tools. One is the one you see here, which is made to be less precise, but returns more results and returns them faster.

The other sounds more like what you are looking for. It returns a more focused set of results and actually generates PR summary reports as PDFs for our customers.

However, we can't really give a public demo to HN of this second one, because it takes 1-3 minutes per run and is quite computationally intensive. (And we were told that our HN launch should demo something the community can easily play around with).

That being said, if you want an example of the output of this second search tool, you can see it here: https://www.linkedin.com/posts/sandy-suh-2ba05bb0_parsagons-...

In short, I don't think you're doing anything wrong. I think you're looking for a more precise version of our product that we weren't able to provide a demo for

40. sand1929 ◴[] No.42323617[source]
Honestly, if you or anyone else has a really easy way to get all communications/publications from every government/government-adjacent body, I genuinely would love to hear it. Especially at the regional/local level, cause those can be a real pain, and we'd love to save ourselves some time