[HN Gopher] Launch HN: Circleback (YC W24) - Tooling to make mee...
___________________________________________________________________
Launch HN: Circleback (YC W24) - Tooling to make meetings more
efficient
Hey HN! We're Ali and Kevin from Circleback
(https://circleback.ai), an AI-powered product to help teams make
more efficient use of meetings. Circleback automatically writes
notes and assigns action items, as well as extract things you care
about after every meeting. Here's a demo:
https://www.youtube.com/watch?v=13NX0QxG5hI&t=10s. 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 and irreversibly delete
all associated data when a meeting or account is deleted and by
default don't store any recordings of meetings. Signup requires an
email and name. Normally we require credit card information but
we're disabling that for HN to make it easier to try out the
product. After signing up, you can: 1) connect your calendar to
have Circleback join your meetings, 2) add Circleback by pasting in
a meeting link, 3) record an in-person meeting from your phone or
laptop, or 4) import the audio/video of a meeting you've previously
recorded (this is what I did with the GitLab meeting). The best
way to see how Circleback works is to try it out in one of your own
meetings. We have a free trial and it takes about 2 minutes to get
started. We'd love to get your feedback and look forward to your
comments!
Author : alihaghani
Score : 45 points
Date : 2024-03-06 17:50 UTC (5 hours ago)
| prakhar897 wrote:
| How does it differ from Otter.ai, or YC funded Cogram?
| candiddevmike wrote:
| Or Supernormal?
| ilrwbwrkhv wrote:
| It doesn't. VC is so inefficient that the same thing has to be
| funded over and over again in the hope one of them win.
| alihaghani wrote:
| The main reason our customers have shared for switching from
| Otter to Circleback is better quality notes and action items as
| well as ease of use. Some of the automations you can create in
| Circleback are also not possible in Otter. I can't speak to
| Cogram as I haven't tried out the product or know much about
| it.
| esafak wrote:
| Show how well it works: embed videos and their diaries in the
| presence of accents, multiple languages, and technical terms.
| alihaghani wrote:
| Definitely, the GitLab meeting linked
| (https://app.circleback.ai/view/lt93jwzdg3hdzcdnxmm) is one
| example. Let me know if any other public meetings come to mind
| I can import to show this.
| candiddevmike wrote:
| It's kind of weird to transcribe a meeting from someone who
| isn't your customer and use it in marketing material (this
| post). Folks who don't have the context would think GitLab is
| a customer of yours and you have their permission for using
| their stuff in marketing.
| teaearlgraycold wrote:
| I would expect them to showcase their dogfooding of the
| product. Just use it for every internal meeting and cherry-
| pick the best result to put on display.
| Kwpolska wrote:
| If you want to test it with multiple languages, perhaps try
| all the parliaments of the world? They tend to have public
| recordings.
| alihaghani wrote:
| Great idea. I imported a Scottish Parliament meeting
| (https://www.youtube.com/watch?v=b2WPUzr5ieI), you can see
| the results here:
| https://app.circleback.ai/view/ltg9d70asxk05j1fyaq.
|
| Right 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.
| snapcaster wrote:
| Interested in this iff it works with non-native english speakers.
| I've found the other solutions I've tried utterly fail at this
| alihaghani wrote:
| We use a Whisper-based model for transcription which works
| really well with accents and supports over 100 languages. Would
| be keen to get your thoughts if you try it out.
| elietoubi wrote:
| Honestly kinda crazy that it gets on top on HN. Goes to show the
| upside of going to YC.
| Kwpolska wrote:
| I believe "getting on top of HN" is a given for those Launch HN
| posts.
| dang wrote:
| As Kwpolska pointed out, Launch HNs are a different category of
| post--they get a guaranteed initial placement on HN's front
| page, though after that they rise or fall with upvotes. They
| never get placed at the top, though.
|
| 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.
| nylonstrung wrote:
| I am so so so sick of these.
| esafak wrote:
| Of what, specifically, and why?
| throwitaway222 wrote:
| I'm guessing it's the creepy "notetaker" that these companies
| have join and listen in on you. It used to be fun to join a
| meeting on time and have a random moment with someone else,
| and discuss something interesting, before the late
| stakeholders arrive.
| esafak wrote:
| So that's a pain point that Circleback is now aware of: you
| want to be able to go off the record.
| alihaghani wrote:
| FWIW, by default we don't store meeting recordings-only
| the transcript. Saving meeting recordings is something
| you can enable from Settings - Account.
| yuck39 wrote:
| Awesome to see you here Ali!
|
| 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!
| alihaghani wrote:
| No way! It's been so long, would love to catch up.
| Kwpolska wrote:
| From the post title, I expected some material improvement to how
| meetings are conducted. Automatically transcribing and
| summarising the meeting does not seem like any serious
| improvement or anything novel.
|
| 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.
| dbish wrote:
| Visual understanding is key to anything that isn't just talking
| heads in a discussion and what almost all of these tools are
| missing, I think because it's a lot harder then using whisper
| and OpenAI (or equivalent) and a lot more expensive/complex to
| ground a conversation in the visually shared data, speaking
| from experience of building a tool that does understanding of
| what you're screen sharing. We focus heavily on this for
| recorded info, pairing it with data on what's actually
| happening on screen (think mouse tracking and action ontology)
| and find it significantly changes the understood summary and
| transcript of any more white collar type presentation or share
| like a developer doing a brownbag (we're not meeting software,
| more Loom-like record and share)
|
| 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.
| cloudking wrote:
| Looks interesting, how does this differ from the built in meeting
| summarization that Zoom AI provides?
| alihaghani wrote:
| The summaries/action items you get with Circleback are better
| in terms of accuracy and detail than with Zoom AI. Running the
| two in parallel for a meeting and comparing the outcomes is a
| great way to test this and how most customers end up switching.
| dbish wrote:
| Is there a metric you use for this comparison? Any reason
| you'd expect Circleback to have context or info that's better
| then Zoom?
| pants2 wrote:
| It will be difficult to differentiate from Otter, Fireflies,
| Sembly, Noty, Krisp, Tl;Dv, Supernormal, Spinach, Fathom,
| Airgram, Noota, Tactiq, Vowel, Jamie, and whatever's built in to
| Zoom, Google Meet, and Teams already.
|
| 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.
| titanomachy wrote:
| > 3. Run it through a basic LLM to correct common transcription
| errors specific to your company's vocabulary and remove filler
| words.
|
| Do you just include a list of common errors in the prompt? Or
| have you trained a model for this?
| pants2 wrote:
| I would love to train a model, but no, it's just common
| errors that I've encountered. It can also fix it in context.
| For example if you're talking about the AI 'Groq' but the
| transcription always says 'grok', then you can, in this step,
| ask the LLM to fix that error if the context is appropriate
| (i.e. it's being used as a noun).
| dbish wrote:
| I do this with just a prompt setting context of the meeting
| if we have it, explaining this is a spoken meeting between n
| people, and asking to correct based on the entire context,
| acutally works decently out of the box.
| titanomachy wrote:
| Damn, I just tried this with ChatGPT (warning: contrived
| example) and it works surprisingly well:
|
| 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.
| pants2 wrote:
| Fantastic example.
| jasondigitized wrote:
| Can you elaborate on the custom prompting? As in "You are a
| scribe attending a meeting of engineers discussing how to
| implement ElasticSearch. Please take this recording and create
| detailed notes of the discussion?"
| pants2 wrote:
| Depending on your use-case it might include things like, "be
| sure to highlight the contributions and points of view of
| each participant", or "come up with a list of action items
| and who's assigned to each one", or "ensure the summary
| doesn't include confidential details about our new product".
| hoten wrote:
| Is this satire? It really reads like that one infamous post on
| the Dropbox HN announcement...
|
| 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.
| pants2 wrote:
| For the record I sync my files using a method similar to that
| user, rather than using Dropbox. I imagine plenty of HN users
| do the same. I'm making this recommendation to other hackers
| on here who don't mind getting their hands dirty.
| dbish wrote:
| Totally agree on the baseline. We've found that adding
| multimodal data like what was onscreen to be a big help to
| improve over this, which is a little more complex. Helps more
| to add action data to like who was typing in what, where the
| mouse was, etc.
|
| 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.
| pants2 wrote:
| Gotta add facial recognition too so the notes can include
| "Bob rolled his eyes slightly when Alice mentioned the new
| reporting procedures."
| Terretta wrote:
| For what it's worth, the new "Meet" add-in from Microsoft for
| Teams is remarkably pragmatic, pulling together transcription,
| speaker tagging, semantic content timeline, meeting assistant
| notes, more, into a common workspace for the meeting.
|
| 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_.
| avsavani wrote:
| Where is tool name Touchbase lol
| nakovet wrote:
| Kudos on launching it, I will be trying in a few meetings to see
| if it goes well, so far I am finding the UI unresponsive, I click
| and it marks an action item as completed (optimistic update) then
| it reverts back to unchecked (failed). I asked the assistant what
| Synergy meant and it just hanged. I liked the use of a fake
| meeting as an introduction to the product and the action items as
| an onboarding checklist, I like that you launched with a delete
| my account as well, most startups take a long time before
| implementing the exit path.
| alihaghani wrote:
| Thanks! We have some kinks to work out with the intro meeting
| and its associated action items. Everything associated with the
| intro meeting is stored client-side hence why action items
| being marked as complete don't persist. You shouldn't notice
| any issues like that with real meetings (let me know if you
| do).
| xwowsersx wrote:
| Congrats on the launch, looks great. How do you compete with the
| built-in Zoom AI companion Meeting Summary? It works very well,
| but it's not even about that. Even if your summary were twice as
| impressive, competing with a tool that's seamlessly integrated
| into users' workflow by default presents a formidable obstacle.
| alihaghani wrote:
| Thanks! Right now, the delta between Zoom AI summaries and ours
| is enough to get many to switch.
|
| 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.
| groby_b wrote:
| A few notes:
|
| # 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.
| alihaghani wrote:
| Appreciate the feedback.
|
| 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.
| vbhakta wrote:
| Congrats on launching, how do y'all differentiate from vowel. We
| used to use vowel but then they went under and we ended up
| writing our own tool, kinda like pants2 mentioned
| aejae wrote:
| How do you compare the pros/cons of having the summarization
| built into the call tool, like with Vowel, vs having more
| control but in separate tools?
| eschneider wrote:
| Is any of your customer data used to train your LLM models?
| alihaghani wrote:
| We don't use customer data to train models.
___________________________________________________________________
(page generated 2024-03-06 23:00 UTC)