Launch HN: Circleback (YC W24) – Tooling to make meetings more efficient
Meetings can suck (https://paulgraham.com/makersschedule.html), but we think better tooling can make them suck less. With Circleback, you don’t have to jot down notes/action items, review past calls to extract what was said, tediously update other sources (i.e. CRMs, issue trackers), or forget where you left off last time.
Ever had to hop on a 30-min call with someone to get unblocked on something you're working on? You go off and do what you need to do, but the outcomes/learnings of that call aren’t documented anywhere. Next time someone wants to do something similar, they end up having a very similar 30-min call. This is a common pattern across organizations, especially medium-to-large ones, and it’s just one example of the vast inefficiency of meetings.
Needless to say, the best meetings are the ones you never attend in the first place. Really good notes let you skip more meetings and catch up with what you need later. In fact, some teams are using Circleback precisely to avoid having unnecessary attendance in meetings. That’s the holy grail as far as we’re concerned!
This space is a good fit for LLMs because while they can’t do everything, they excel at processing unstructured data like a transcript. To get rid of even 20% of busywork around meetings is already a massive win—and we think the potential is a lot higher than that.
For a realistic example, I imported one of GitLab’s meetings from their YouTube channel (they famously publish them!) into Circleback. The original video is here https://www.youtube.com/watch?v=1dwzE5j-IQw (though you won’t want to watch the whole thing) and you can view the notes Circleback wrote for it here: https://app.circleback.ai/view/lt93jwzdg3hdzcdnxmm.
You can create automations in Circleback for tasks that come up repeatedly. For example, you can create a workflow to identify feature requests that come up in customer interviews and create tasks for them in your issue tracker and post an update to Slack with the notes from the meeting.
Some of the ways people currently use Circleback include: remote teams gathering context from meetings they didn't attend; sales teams automatically updating their CRM and keeping track of what they last talked about with a client; and executive assistants sharing meeting minutes with the rest of the team.
For engineers, past meetings turn into a knowledge base, so you can search for anything that came up in a meeting before—e.g. some obscure quirk about a sequence of API calls, or how you figured out some bug 6 months ago. If you had a meeting about how to run a service, you can use search to find that meeting and watch the exact moment it was shown. Also, some engineers appreciate having issues be automatically created for tasks that come up during standups. That may not be how everyone likes to work, but it beats doing it manually.
With regards to privacy/security, we use industry-leading practices for handling customer data (i.e. secured at-rest and in-transit) and we're in the process of getting our SOC 2 type II certification. https://security.circleback.ai has more on how data is handled. We also immediately...
97 comments
[ 3.2 ms ] story [ 154 ms ] threadRight now, speaker identification (and thus everything downstream that depends on that) works better for meetings where Circleback joins the call (compared to those imported) but I wanted to share this anyway.
I think the thing I'd like most is if the AI could "cite" its source (i.e. provide a timestamp of when exactly the statement was made that it's summarizing) directly inline in the summary (vs. only in the transcript).
Similarly to how they're used in AI-powered search (e.g. Perplexity), inline citations would help me easily cross-reference statements and gain confidence that the summary is accurate and the information can be relied on.
Launch HNs are one of the things that HN gives back to YC in exchange for funding it. The other big one is job ads for YC companies. This is in the FAQ: https://news.ycombinator.com/newsfaq.html.
You may or may not remember, but we used to run similar tech YouTube channels as young teenagers and worked on a few videos together. I've been following circleback since you guys were accepted into W24. Congrats on the launch!
Also, it seems to be focused on what is said in the meeting. But what if the meeting is heavy on the visuals (a PowerPoint presentation, a demo) and the spoken content doesn’t make much sense without seeing the shared content? I imagine the output will not be very useful in that case.
Circleback folks, cool product, if you want to add visual/screen shared data as a dimension of understanding for you hit me up (diamond@augmend.com). We’re setting up a service for a few meeting/recording understanding products and would love to help you out here too.
Meetings, emails, and other mundane stuff like documentation generation and such would be really prime things I'd love to see tackled. Sadly, its not sexy enough to expend resources on unless you throw in some AI or other trendy hook.
Email in particular is the bain of my existence. Our tools around it suck (Outlook), provide no real management features (gmail) or seem to be stuck in the early 2000s (Thunderbird).
Particularly I am desperate for more developer friendly ways to script, automate and manage email correspondence and chains because I feel 80% of my day is wasted just reviewing wtf is in my inbox.
However. "meetings rely on visuals sometimes instead of just spoken words" doesn't feel like a fundamental issue that breaks this thing. It just sounds like a feature request.
I agree that there's more that can be done for meetings that heavily rely on visuals. Right now, we save what was screen shared (if recordings are enabled) but the visual component isn't used in generating outcomes. Lots of interesting things that can be done with multimodal models here.
I evaluated many of these for my company and ultimately found that they were pretty much all bad due to lack of customizability. I would suggest anyone here looking for an AI meeting note taker to write their own:
1. Pull down the raw recording (using built-in recording or Recall.ai).
2. Run it through Whisper or Deepgram with a custom prompt and custom vocabulary.
3. Run it through a basic LLM to correct common transcription errors specific to your company's vocabulary and remove filler words.
4. Run that through GPT-4 or your favorite powerful LLM to generate notes, iterating on the prompt to tailor it to your industry and meeting style.
All of my coworkers agreed that the custom solution (which took me a day to slap together and is close to free) is dramatically better than any of the off-the-shelf meeting note takers we have used, no contest. Being able to customize the transcription to capture industry-specific terms, acronyms, or internal codenames, is an absolute must.
Do you just include a list of common errors in the prompt? Or have you trained a model for this?
PROMPT:
The following is a partial transcription of a meeting at a tech company. Please correct any transcription errors, using clues from context as well as the fact that this is a technical meeting.
Person A: we shouldn't use a sequel database to store this data, we probably want to use something Callum nor instead since it's more efficient than row-based storage. Something like a patchy park or arrow.
Person B: What's park, eh?
Person A: It's just a column-based storage format with efficient Jesus or snappy compression.
RESPONSE:
Person A: "we shouldn't use a SQL database to store this data, we probably want to use something columnar instead since it's more efficient than row-based storage. Something like Apache Parquet or Arrow."
Person B: "What's Parquet?"
Person A: "It's just a column-based storage format with efficient GZIP or Snappy compression."
EDIT: mistral 7B on the other hand botched this terribly, and made up a lot of nonsense unrelated to the transcript.
https://news.ycombinator.com/item?id=9224
For every one person who'd want to do it the better DIY way, there are 10000s that can't (or won't), but would still find value in the service.
nonetheless, there are ppl whose needs do not exceed their skills
I’ve also been playing with pulling in knowledge base context or reading relevant web pages for unique words to create that initial prompt and custom vocab automatically.
People are just discovering it this past quarter or so, but when they see the scope and utility of what it can do, they start recording everything.
https://support.microsoft.com/en-us/office/stay-on-top-of-me...
Right now, we provide the following customization levers: custom vocabulary (configurable from Settings → Account) and custom prompts (via insights you can define in workflows). We're also working on adding more control over the verbosity of the notes.
If you're looking to have full control over the outcomes generated, I would also suggest building your solution if the building/maintenance costs make sense. We're focused on providing an out-of-the-box solution that works really well with minimal setup (with customizations available for power users) and providing value beyond just summarization with search, workflow automation, and collaboration features (i.e. sharing, commenting) for teams.
If you do decide to try out Circleback, I'd love to get your thoughts as someone who's very well-informed about this space!
An engineer who can slap together a good meeting transcriber in a day is probably paid $2,000 a day or more. The opportunity cost of having them work on a random project is far more than the cost of their salary. The ongoing maintenance cost of the custom meeting transcriber is probably 5-10 engineer days over its lifetime. The added utility from slightly-better meeting transcripts does not compensate for the staggering engineering cost of the custom solution.
Circleback and other meeting transcribers will surely get better and surpass the quality of the custom transcriber. Then the team must spend more time switching to Circleback and deleting the custom transcriber.
However there's lots of manual steps here and it would definitely be more convenient to have a tool that just does all that straight away. Also, Whisper unfortunately does not (yet?) support diarisation (identifying which speaker is talking).
Anyways, those are the key takeaways from your data driven thought leader, yours truly.
We've also built (and are continuing to build) functionality that goes beyond summarization:
- Integrations that connect with other apps/CRMs to automatically push relevant data after every meeting.
- We can automatically identify and take action on insights you care about. An example I showed in the linked video (https://youtu.be/13NX0QxG5hI?feature=shared&t=144): I have a workflow to identify the customer's industry after every demo call I do. There are some really cool use cases here, especially when combined with integrations, i.e. automatically updating a field in a CRM with the customer's industry, or create a new row in a Notion database. We'll soon be adding schema support for insights so you can have more control over the structure of the insights generated, too.
- AI-powered search across all your meetings: being able to get an answer to a question using the context of your/your team's meetings (regardless of whether they happened on Zoom, Meet, Slack huddles, or in-person) is quite powerful and something we're working on making much better.
# your pitch above targets the wrong audience. People on a maker schedule need to eliminate meetings, not have better transcripts. People on a manager schedule are the ones that need help, and "meetings suck" isn't talking to them.
# The places that will be really interested are places with lots of meetings. I.e. mid-to-large enterprise. Neither your pricing nor your marketing targets them.
# way too verbatim. Most meetings don't require that level of detail.
# You might want to link notes into the transcript/video for the parts where people want more detail.
We've had some customers request more detailed notes which we've enabled for them on the backend. We have plans to release more self-serve customization around verbosity as there's no one-size-fits-all solution.
Agree on linking notes to the transcript. Citing sources (from the transcript/recording) is something we're introducing to more surface areas in the product.
We actually built some hubspot and slack integrations and you folks released those same integrations soon after. It proves that you listen to your customers and develop meaningful features. Kudos to you!
We’ll keep recommending you over any of your competitors because your output is so much better!
Wish you the best and good luck!
What kind of marketing trickery is this?
Serious question.
it's called the "circleback trick"
We learned this by watching how the community reacts to Launch HN posts. If it feels like the company isn't "ready", the reactions tend to be skeptical and disinterested. "Ready" for a Launch HN can still be early-stage, but the company needs to feel real, there has to be a real product, more than a bare-bones website, should show pricing, something more than "Book a demo" for trying out the product, and so on.
By the time a company has developed enough to check all those boxes, it's natural that it would already have a bunch of users, as Circleback does in this case. In fact, that's the ideal case—the startup should be making something people want, and have users to back that up, before doing a Launch HN. (There are always exceptions, like if you're developing an electric airplane, a cancer therapy, or something else that takes years before users can try it.)
After puzzling over this, I came to the conclusion that the bar is higher for Launch HNs because they have the official stamp of (1) being YC-funded and (2) getting placed on HN's front page. (For background on the latter, see https://news.ycombinator.com/newsfaq.html and https://news.ycombinator.com/yli.html). That raises community expectations, and rightly so!
> Needless to say, the best meetings are the ones you never attend in the first place.
The best (avoided) meetings are the ones that are pointless and *are cancelled before they happen*.
You've done me one better in this respect, point taken!
Edit: for those curious about how these posts get made—I work with the founders to try to figure out how to talk about their startup in a way that might appeal to HN. See https://news.ycombinator.com/yli.html for the instructions I send to YC founders.
It's not possible to get this perfect, but we can usually avoid obvious pitfalls. With a startup like Circleback, for example, I felt it would be important that the text not feel like it was written primarily for managers. Hence the pg essay link as well. (I might have pushed this too far, though, because https://news.ycombinator.com/item?id=39621356 makes some good points.)
It's not a question of making anything up, of course—the most important thing is that everything be true. But that still leaves a lot of range about what points to emphasize, and there I try to be useful to founders, especially founders who haven't spent countless hours on HN and thus don't have an instinct for the culture here.