5 ms·

Show HN: Prosona – Your co-pilot for intelligent responses in Slack

13 points by hgunasekara 3y ago

Prosona was built to solve a personal problem at work - we’re constantly bombarded by questions from colleagues and our productivity suffered as a result. All knowledge workers spend 15% - 20% of their time searching and gathering information. This often occurs when a colleague asks for a status update or how something works. Although the answers may be simple, the process of referring to Jira, grabbing links, and formulating an apt response is time-consuming. Until Prosona, this inconvenient context switch was just an accepted aspect of work life. The issue is compounded by the dispersion of knowledge across a variety of apps such as Google Drive, Slack, and Email. We lose track of where information resides, or worse, forget about the existence of some information within our tools altogether. Adding new information to the mix becomes a tedious task of properly categorizing and documenting it. This is where Prosona comes in. Our system integrates with all your knowledge stores, learns your unique tone of voice to then generate a draft allowing you to further personalize the message if needed. In terms of security, we place the utmost importance on the protection of your data which is why it is encrypted in our vector DB. It can only be decrypted by you. As an employee, your focus should be on achieving what truly matters: whether it's meeting your OKRs, making an impact, or enjoying pastimes outside work. The last thing you need is to get bogged down in crafting routine messages to keep colleagues updated. All you have to do is type `/prosona` in Slack, and a beautiful draft will be written grounded in your context and tone. Today we’re launching our private beta. To take part, sign up on our website. We're providing access to the first 500 people who register.

borgiancalendar 3y ago
How is this different from Glean?
- hgunasekara 3y ago
  We are integrated directly into your workflow, helping you answer questions that come to you. Even though we had Glean at our companies, we still got faced with many questions. Prosona saves you time compared to doing a search on Glean yourself since it is also written in your tone.
smt88 3y ago
> we place the utmost importance on the protection of your data which is why it is encrypted in our vector DB. It can only be decrypted by you You need to explain this in a lot more detail. It sounds essentially impossible to provide this service with zero-knowledge encryption, but I'm open to being wrong about that.
- hgunasekara 3y ago
  Hey smt88, we use a similar approach to Cloak AI which is encrypting vector embeddings. https://ironcorelabs.com/blog/2023/announcing-encrypted-ai-embeddings/ https://ironcorelabs.com/blog/2023/announcing-encrypted-ai-e...
  - smt88 3y ago
    Do you have more detail? Like a lot more? And are you being regularly audited by a third party? Saying "we use a similar approach to Cloak AI" isn't good enough for many reasons. They don't provide a lot of detail to begin with, so it's unclear their approach is proven. And you're not even using their approach, just a similar one. Even if we had a long whitepaper about how you do things, that's still not good enough because we don't know that you're actually doing it. That's why the audit is important.
    - hgunasekara 3y ago
      We want to emphasize how important privacy is to us. It felt wrong to be able to access a company's private knowledge so we set out to make the system as secure as possible. In terms of our embedded vector DB, we use Homomorphic Encryption to vectorize the embeddings. We are still super early - started hacking on this project 7 weeks ago, in the near future we plan to get SOC 2 compliant so we can definitively prove that we are secure. You can read more about Homomorphic Encryption here: https://aclanthology.org/2022.finnlp-1.4.pdf https://aclanthology.org/2022.finnlp-1.4.pdf
pavel_lishin 3y ago
> This is where Prosona comes in. Our system integrates with all your knowledge stores, learns your unique tone of voice to then generate a draft allowing you to further personalize the message if needed. Is there an option to make it clear that the response was written not by me, but by an AI assistant?
- hgunasekara 3y ago
  Not for our MVP but we are working on a feature that will all you to show whether your response was AI assisted or not
  - pavel_lishin 3y ago
    As someone who receives slacks, I'd also love for this not to be optional. I don't want to talk to someone IRL and have them look at me blankly when I describe something they allegedly wrote to me.
    - hgunasekara 3y ago
      We will definitely make this a feature! Providing awareness on whether it is an AI or human response will be key for trust with colleagues. It is worth noting that messages are approved before being sent so you have total visibility as to what is going out.
xena 3y ago
This feels absolutely dystopian. I hate how things like this are almost guaranteed to be ubiquitous and how it will encode corporatespeak into the fabric of how even small companies make decisions. Ultimately though, unless I can run it on my own machine, it is not truly private.
- hgunasekara 3y ago
  Hey Xena, we fine-tune on your personal tone - this makes responses sound authentic ad is a great base to make further edits. We did this to amplify the individual since we believe people are the center of knowledge sharing. Discussion is key and we felt documentation alone wasn't created that space to share thoughts and ideas.
  - xena 3y ago
    Are you looking for people to do a review of your production environment for security? Please email the address in my Hacker News profile if so.
ulrashida 3y ago
Are people truly this lazy? What happened to knowing how to answer questions related to your job? Finding a status update in your area should take seconds -- if it doesn't it suggests you don't have adequate systems in place which would then in turn would also kneecap an AI assistant. How bizarre.
- hgunasekara 3y ago
  Think about it like a bike for workplace communication. While walking is fine, you can get further with a bicycle. We are streamlining your workflow to give you back time - this allows you to focus on other, more important, things
stitched2gethr 3y ago
This may be the wrong way around. If the example use case is that a colleague asks how something works and I respond via Slack with Prosona, shouldn't they have just asked Prosona or something like it? Why am I in the loop now? If they need a status update, how will Prosona have access to information which is not in my head, but also not accessible to my colleague? I do appreciate exploration of AI in different settings, but I feel like any tool that only works in one of my screens isn't going to be what wins 10 years down the line.
- hgunasekara 3y ago
  Great question! We realized documentation will always be patchy and imperfect which is why we put you in the loop so that you can provide extra context. There will always be information in your head that is not written down and we streamline your ability to disseminate that information. It also means there is accountability in the case where the question is used to make a decision. This launch is only the first step for us. In the future, edits you make to the response could be used to update upstream documents making the hive mind even more connected.

13 points by hgunasekara 3y ago

borgiancalendar 3y ago

How is this different from Glean?

hgunasekara 3y ago
We are integrated directly into your workflow, helping you answer questions that come to you. Even though we had Glean at our companies, we still got faced with many questions. Prosona saves you time compared to doing a search on Glean yourself since it is also written in your tone.

smt88 3y ago

> we place the utmost importance on the protection of your data which is why it is encrypted in our vector DB. It can only be decrypted by you You need to explain this in a lot more detail. It sounds essentially impossible to provide this service with zero-knowledge encryption, but I'm open to being wrong about that.

hgunasekara 3y ago
Hey smt88, we use a similar approach to Cloak AI which is encrypting vector embeddings. https://ironcorelabs.com/blog/2023/announcing-encrypted-ai-embeddings/ https://ironcorelabs.com/blog/2023/announcing-encrypted-ai-e...
- smt88 3y ago
  Do you have more detail? Like a lot more? And are you being regularly audited by a third party? Saying "we use a similar approach to Cloak AI" isn't good enough for many reasons. They don't provide a lot of detail to begin with, so it's unclear their approach is proven. And you're not even using their approach, just a similar one. Even if we had a long whitepaper about how you do things, that's still not good enough because we don't know that you're actually doing it. That's why the audit is important.
  - hgunasekara 3y ago
    We want to emphasize how important privacy is to us. It felt wrong to be able to access a company's private knowledge so we set out to make the system as secure as possible. In terms of our embedded vector DB, we use Homomorphic Encryption to vectorize the embeddings. We are still super early - started hacking on this project 7 weeks ago, in the near future we plan to get SOC 2 compliant so we can definitively prove that we are secure. You can read more about Homomorphic Encryption here: https://aclanthology.org/2022.finnlp-1.4.pdf https://aclanthology.org/2022.finnlp-1.4.pdf

pavel_lishin 3y ago

> This is where Prosona comes in. Our system integrates with all your knowledge stores, learns your unique tone of voice to then generate a draft allowing you to further personalize the message if needed. Is there an option to make it clear that the response was written not by me, but by an AI assistant?

hgunasekara 3y ago
Not for our MVP but we are working on a feature that will all you to show whether your response was AI assisted or not
- pavel_lishin 3y ago
  As someone who receives slacks, I'd also love for this not to be optional. I don't want to talk to someone IRL and have them look at me blankly when I describe something they allegedly wrote to me.
  - hgunasekara 3y ago
    We will definitely make this a feature! Providing awareness on whether it is an AI or human response will be key for trust with colleagues. It is worth noting that messages are approved before being sent so you have total visibility as to what is going out.

xena 3y ago

This feels absolutely dystopian. I hate how things like this are almost guaranteed to be ubiquitous and how it will encode corporatespeak into the fabric of how even small companies make decisions. Ultimately though, unless I can run it on my own machine, it is not truly private.

hgunasekara 3y ago
Hey Xena, we fine-tune on your personal tone - this makes responses sound authentic ad is a great base to make further edits. We did this to amplify the individual since we believe people are the center of knowledge sharing. Discussion is key and we felt documentation alone wasn't created that space to share thoughts and ideas.
- xena 3y ago
  Are you looking for people to do a review of your production environment for security? Please email the address in my Hacker News profile if so.

ulrashida 3y ago

Are people truly this lazy? What happened to knowing how to answer questions related to your job? Finding a status update in your area should take seconds -- if it doesn't it suggests you don't have adequate systems in place which would then in turn would also kneecap an AI assistant. How bizarre.

hgunasekara 3y ago
Think about it like a bike for workplace communication. While walking is fine, you can get further with a bicycle. We are streamlining your workflow to give you back time - this allows you to focus on other, more important, things

stitched2gethr 3y ago

This may be the wrong way around. If the example use case is that a colleague asks how something works and I respond via Slack with Prosona, shouldn't they have just asked Prosona or something like it? Why am I in the loop now? If they need a status update, how will Prosona have access to information which is not in my head, but also not accessible to my colleague? I do appreciate exploration of AI in different settings, but I feel like any tool that only works in one of my screens isn't going to be what wins 10 years down the line.

hgunasekara 3y ago
Great question! We realized documentation will always be patchy and imperfect which is why we put you in the loop so that you can provide extra context. There will always be information in your head that is not written down and we streamline your ability to disseminate that information. It also means there is accountability in the case where the question is used to make a decision. This launch is only the first step for us. In the future, edits you make to the response could be used to update upstream documents making the hive mind even more connected.