Vinton Feature Spotlight: Prompt Templates Deep Dive

Is showtime. Now we’re live. Hello, Luca. How are you today? I’m doing very well. How are you? I’m good. Thank you for asking. Very good to see you. Hello, everybody, and thank you for joining us. For yet another LinkedIn live session. LinkedIn live session. This month is yet another new thing. A very interesting, small but mighty feature.

Yeah. So what what’s the focus for today? And what are we actually let’s start with what are we doing here? Who are we? We are eventing by native video. We definitely are. And we are here so we can talk to you guys more about what we what we do, why we do it, the motivation behind it. If there are any questions that we you guys would have.

So this is more of an informal session where we go through different features that we have because we, realize that we don’t usually get that time. So this is, the perfect platform for us to share with you guys and get feedback from you guys. Yeah, definitely. We tend to communicate a lot about the ecosystem. We’re very interested in everything that is happening in the ecosystem.

AI is a big topic for us. And this session in particular, once a month is dedicated to introspectively. You get questions, tell a bit more about our our system. Our solution is a conversation intelligence solution native to Salesforce, which means that all conversations or video calls, phone calls, in-person conversation, and even voice notes can be recorded, transcribed, and summarized directly in Salesforce within the context of the records, the objects, the processes that already in systems.

So that’s what we do. That’s what we are. As part of that, there’s a lot of things that we do from a technology perspective that are really super cool, and Mina knows one thing or two about it. So why don’t we go right into so what’s the topic for today? While I set up the. So the topic for the day is Prompts and how we offer different solutions and how you can prompt the output that you would get from a call or a meeting, or an audio note or a voice note.

So because because the AI needs instruction instructions, right. So what exactly. So we in our system have designed various IT, have given users flexibility on how they want to design. The instruction of course is out of the box instructions that come with the solution. But then there’s an added layer of you being able to as a user, being able to define how the output should look like in different scenarios.

Correct? Right. And we’re actually already right into, the, the interface. This is one of the key, elements of and components of our solution, which is what we call the Vinton Library. And you can see this is a user’s library. So all my activity gets, you know, added into this library. Obviously, this is, demo.

Actually, this is a dev environment even. Right. So we’re we’re going really behind the scenes here. And the reason why we’re going behind the scenes is because the first feature that we’re going to show is in the admin panel. It’s hidden behind. Inside the internal we go to, I promise. And here it is. So if you have internal you will not see this one yet.

Because this is something that we are, for which we are finalizing, the, the testing and then, you know, ready to, to deploy. So what is the system prompt? So why am I so excited about it? So system prompt is basically a prompt that would go before any input that you give to AI. So for for example, you have a company and you want to define behaviors and roles.

Some things are genetic and you don’t want to define them over and over again. And every individual instruction. Yeah. So it could be the behaviors. It could be the structure of the output, the tone that you would want in your follow up emails for example. So there are a lot of things that can be defined on a system level.

And they apply to it. Exactly. And then yeah. Yeah. Which industry you’re from is exactly the example. The example was just what you said. Right. So the, the client asked about this and we thought it was a good idea. And there’s specific industry is, pharma and there’s a number of acronyms and number of, definitions.

Even the style of communicating and creating information when, you know, it goes into the CRM, for instance, there is very specific to that industry. And so framing or pre framing, the, the ecosystem within which that instead the particular company is operating, informs upfront, everything that follows every, all the other instructions already incorporate that.

Exactly. And we’ve created an example here. Yes. So you can see that in our example we have a branding rule defined. Maybe Luca would want to show. Yeah. So around well, this is, you know, a very basic example. But you know, one problem that system prompt is solving for us is that we used to be called native video.

The company actually, that Vinton operates from and belongs to is, still called Native Video Limited, based here in London in the UK. But around a year ago we decided to go to the market with our product name, Vinton and Vinton became the name of the company? Yes, and everybody used to call us, but we didn’t and nobody knew native Video it’s not exactly.

So now, you know, we we are known as Vinton, but, some of our emails are still, you know, sent, to Nativevideo, the calendar and calendar invites sometimes are, you know, Nativevideo, sometimes are Vinton because, you know, with the domain. So anyway, this generates confusion. And sometimes, you know, Vinton will will say the meeting was attended by Luca from Nativevideo, and I don’t really like that.
I love it. Right. So it’s obviously still. But. So you get what I mean? So the instruction reads very simply, there’s a branding rule that applies to everything that we do because you are an AI processing, you know, engine for Vinton. And so we give it an identity. And then through this we say always refer to the company as Vinton in outputs, even source material mentioned in the video.

So, you know, as I mentioned, the domain name of my meeting invite, replace it with Vinton. Unless there is a specific reference to the underlying company or the underlying, you know, what is it like for plus? That’s a plus for now. The operational framework as well. We added so this is more like the industry indication, right.

So you say, look, you know, this is where this company operates. So you expect things to be relating typically to the Salesforce ecosystem, the enterprise Salesforce ecosystem. Your role is to analyze. You know we give it a role. We understand we tell it to that is a very, you know, very much an expert on so subjects fields pipeline.

Some you know, we make references to, qualification and frameworks for, for sales for other, situations that, Vinton finds. Then we define the scope. Yeah. So again, these things, if you go back to the, the, guidelines that you showed, you can see that these things are the things that you might not add in the individual prompt because you’d be like, okay, when you’re creating a prompt or when we’re creating problems with our customers, they would be like, okay, summary I need somebody to be bullet points.

I needed to have a chronological order of what happened in the call. If there were questions, I need the questions to be generated. So whenever we’re defining a prompt for a particular, template, for example, we just focus on the call and what output we need from the call. But in all of those prompts we don’t add the guidelines.

So you would want to add for your organization because they’re already here. So now what would happen is that every prompt input they you that would go to our system, it would have a system prompt appended for it. And then the insights would be generated. Taking that into consideration, this would not be a part of output but the output when it’s being computed, it would take in consideration of all the rules that you have.

Yeah. This could be acronyms that maybe you use in your organization. I know that a lot of sales organization have their own acronyms that they define within. Yeah. You know, that’s organization specific. So you can just say, okay, if a user uses this, then it means that, yeah. Which, which is a very, very interesting approach is not it’s not new.

We haven’t invented it. Right. So it exists as and it’s the imagine it as an umbrella, prompt. Right. So something that always is set as a framework upfront for any specific instruction. And I think this is, a good example. You if you’re following this and you’re familiar with with with Vinton, then you know what we’re going to talk about next, which is templates.

And you will see how these actually precedes and it’s a prefix that is added before then, you know, the templates process, commences. And then ultimately it informs the rest of the outputs, that are generated through binned. Correct. So I’m, I’m really in love with it because I think, especially for some of our clients that are operating in very niche, sub industries, this will be very relevant.

It will inform a lot of, you know, up front a lot of the things that otherwise they would need to specify to a level of detail that they don’t want to do at every single step of creating process. Right. So when I perform. So I’ll just go over how we’ve designed the AI prompts. The first tab you can see is default prompts, which means that, oh this one.

Yeah. Okay. Okay. Well, we’re doing it so you can see that we have different types of calls. We have voice notes, video calls, in-person meeting phone calls and Scribe. So you can define the prompts for each of them. You can define for example in video call you can define how the output would look like in a summary, how the follow up would be structured.

Next steps highlights, coaching everything. You have full customization over how you would want it to, appear after the call is processed. First we have default prompt. So the way it works is if you’re a small organization, you probably not don’t need a lot of templates. So you can use just default prompt. But if you are a medium size organization then you would have different departments for example.

And in different departments you have different type of calls. Then how would you structure those goals? To solve that problem we created templates, which means that you can define the context of the call. So first the system would check the should we go into what which one is your favorite? I think we should go through the first one.

First one. So this is an example that is very similar to a qualification. And so the first, client call and yeah, you were you were saying, I think this is a very interesting, sequential, kind of a correct, set of events that happens. Right. So the first will be now the system prompts. So the persistent problem is the first information that is given to the AI and says, look, this is the framework of reference.

And then he goes into this new step which is different from the default prompt. Because this what does it do. So this is a template that you define. Now if you have ten different types of calls in your sales processes, the first you would have an introduction call. Then you would have a demo call. And then afterwards you’d have a call in a later stage where in each of these calls you would want different output flow for yourself from the call.

The summary would be valuable to you if you have certain type of output. I know that you use medic and band band for the initial call and then metric for the later stages in a sales call, but if you want the output to be, framed on those frameworks, then we would need to have a template that apply this, for instance, because the instructions are different exact based on.

While this is an early stage, a client facing meeting between two organizations exploring potential partnership or collaboration. So so obviously this is a generic and the context may not create very in our clients, in collaboration with me not typically create very specific context where to define exactly what meaning was seen. Right. So this is if the context is a sales early stage conversation client facing such and such, then these are your specific instructions.

So if we, you know, come to the video call prompt and we go to the summaries, you can see that, you know there is a draft and it’s a draft because this is not taken. You know, if you’re familiar with with Vinton, you know this already. And this is a very specific set of instruction that is different from the default, which is a bit more generic, which supplies, if but if the context is match, then this will be the structure.

And you can see it’s very it’s very specific, to the the introduction I think where, you know, calling it introduction and cool. And again, you probably can think immediately of how this would apply to your own organization and how you’ve defined those frameworks of reference. Right. So for discovery, call what you ask what is important that gets captured and added into the CRM from a coaching perspective.

Equally, what is important to refer to in terms of was this a good call? Was this a bad call? In relation to an introductory call? Yeah. So I mean, again, it can be customized to the level of detail that you would want it customized. Do we have with our customers, we have created problems that create reports.

After a call because the nature of the conversation would be, such that after the call, for example, if I have a call with you, we would talk about a report that we have to create. And after that call, I would need to create that report. We also created problems to the level of detail, which would have everything that I would need in the report.

It’s just the structure. If we know that this is the structure of the call, that would happen, then we can define a context for it, and then we can have a certain output from that particular call. That’s that’s absolutely fine. And so this actually is and can be design specific for the different types of recordings that are typically captured via Vinton.

Right. So the audio note prompt is the classic voice note as I’m walking out of a meeting and I’m like, I have, you know, everything fresh in my head. I just, you know, and I’m, I’m making the gesture as if I’m doing it on, on mobile. But obviously it’s also, and maybe we should show an example of a voice note, but then video call is the classic, you know, teams meeting.

Zoom meeting. Or zoom, of course. Zoom meet and zoom and me know an image or even a, a huddle is like, people talking. You know what? What is the context of the conversation? What are the specific instructions? So if we go back to is console. So if we go back here you will see that, you know we have different contexts.

Again. You know, we have organizations that have very long list. They have, you know, 3 or 4 different context in there for specific instructions for sales, customer success, internal calls. You are saying, right. So we could actually, meet as colleagues and strategize about a product or strategize about a client and that, you know, would require a different approach from a summarization perspective, from next steps.

And, you know, the level of detail, the format, all of that can be prompted. Right? I mean, an internal meeting that I have with the team would have a different follow up email and a meeting with that I have with with you would have a different follow up email because then you would be looking for just bullet points, because obviously you don’t have the time to go to the details.

But if I have the meeting with development team, they would need details about what needs to be implemented, how it needs to be done. So again, you can, formal, you can create different plans based on the situations that you have in a particular meeting. So then those frameworks, of reference, those, those instructions, the templates are then applied.

Then we can see an example here. Let me zoom in. Client introduction call. Right. So this was identified, by by Vinton as matching the context that was at the start of the instruction. So the instructions applied reflect the specific instructions that are in this. Now, these are the templates. You know, we also have the ability, obviously, to force, a new language.

And you can see how they appear in front of the user again. The user typically will not have to worry about this because, Vinton will understand the instruction and apply the correct, framework and the correct template. But in case, you know, we can always look at the same conversation through a different lens or different set of lines, for instance, that that would be, how, these are, applies.

This is very, relevant for our clients because as you were saying, out of this, they produced the very specific set of information that they want to capture across the entirety of the records and their objects in their Salesforce. Now, I think this is clear. I want to come back here because I really like the way these work into AI prompts and templates.

Now, do we have questions about templates? Do we have questions about my new favorite? The prompt? Which, you know, system prompt? I’m very excited about it. Just because it’s there, I feel, it’s a such a simple, approach. But I think one that can make it through difference. And as I said, especially for, for, you know, those clients that, you know already, you know, here and they’re already using it, they, they really feel like it’s making a difference, you know, across the board, the type of information and the, the area.

When I presented this, the other day, a comment that I got was, oh, but you’re you’re actually giving us access to the true backbone of your system. And, you know, my reply was, well, it’s not our system is your system and your sales force. It’s your data you should be able to extract from your transcripts, your conversations your voice notes, exactly the information in the format and the length and the frame of reference that is relevant for you. And I think this is probably the surprise there was because other systems don’t allow that. They’re not as open and as flexible. But I think for us, because we operate in different industries and different types of industries use us, there’s no one formula that we can use would fit everyone. We work with a lot of our clients and customers to make sure that we generate prompts that they need. So if you are thinking of onboarding with us, we get full support, on generating and drafting the problem. So we work together with all of our customers to make sure we trust the prompts to generate the output that they need in their CRM.

So we don’t leave people alone. No exactly, exactly. And because we are actually a very interesting exercise, that we see with our clients because they, they truly warm up to the idea of, you know, okay, start with the default. Oh, yeah. This is interesting. It’s probably 80% where I need to be exact, sometimes 70, sometimes 90.

Right. But and then they say, oh, but if we tweak this or if we make this a bit longer, we go a bit more in detail in this area. And we define a, you know, completely different approach for this other way. And that is an iterative process that, you know, kind of as time goes by, they internalize and then, you know, autonomously.

So we just check in every once in a while. Hey, are you guys doing, you know, the create a new template lately or how are templates going? I think this the recent past few, past few months or past six months is where the templates have gained a lot of traction, where everybody is, going. Yes, we need customized templates.

We need to customize it to the level of detail that we need in our CRM. We didn’t see this before. So this was not something that happened before. Like if you go six months ago, but people, customers didn’t come back to us and ask for specific format and summary or specific format and follow up email. But recently this has been a very important requirement for all of our customers, and that’s why we worked in this direction as well.

Last time we talked about the follow up email and how you can we we generate it. And this time we’ve worked on, system problems. So because if our customers are using something, we want to make sure that it is to the level of the standard that they would want it to have. Yeah. And we’ve tested it internally and it’s really helpful.

It is the system prompt is you already mentioned what, is interesting. You know, your observation, I think, is, is determined by the fact that more and more individuals like us are, becoming familiar and not scared about the prompts. You know, when they interact with their GPT, they understand that, you know, their approach needs to be contextualized, that they’re approaching it.

And that’s, you know, they see the quality of the output improving. And then the other side is that we’ve seen more and more of our clients investing in bringing on board AI expertise. And I mean, this is ground level, for a lot of them, right? So you have a lot of data that come all into your Salesforce from all the conversations across the organization.

What do you do with that? How do you capture how do you make the most of it and how do you you save and surface the right information at the right time that is relevant for the business. And exactly that’s customized. Yeah. So to get to that, to the AI engineers who would come in and need that information, the first thing that would they would need is the data to be classified the way they want to classify.

So that’s the first step of how we get to that point, because now we’re seeing our clients building AI pipelines and how they can use these call data to get transcript, to get insight beyond just one call, to get insights, beyond just what happened in one conversation. So that’s also something that the that everybody is going in the direction of.

And I think internally, we decided that we need to make sure that the input should be perfect. And that’s how the output would be perfect as well. So that’s just a step towards that. And this would and I think that’s it. So we don’t have any questions. So we answer all your questions. Well, if you have more questions, maybe leave them in the comment.

If you’re not watching live, you know, there will be comments still. And we will monitor that for the next few weeks. This was super interesting. I’m really, excited, about the the adoption of the prompts in our system evolving to support supporting more and more ways of customizing and personalizing, the, the solution that we provide to our clients.

So was done was, you know, done with an team, here at Vinton and, that’s good. But, yeah, I think that’s that’s it for today. Thank you so much for joining us or for watching this. If you’re watching this after, thank you. And it’s worth 25 minutes. So yeah, you and 24, you know you’ve done well if you’re still watching.

Thank you. We appreciate you. And we’ll see you in what a month right. And a month. Yeah. And that’s be fully yeah. That’s going to be another one. Don’t miss the next one.