Clip
Summary
Jecelyn showed a pipeline she's building to turn Evo Weekly recordings into email digests, blog posts, and per-feature video clips. Researcher fetches the transcript via custom connectors, a skill drafts the structured article JSON, and Workflow Automation crops the video and screenshots. A custom MFE editor lets reviewers adjust and reorder features before publishing. Built entirely on Evo pillars — Researcher, Connectors, Workflow Automation, and Evo Builder.
Screenshots
Transcript
Can you see my screen?
Elliot Iles 2:59
Can now?
Jecelyn Yeen 3:00
Yeah, so I want to say that the work is currently a work in progress, but I would say I would just share like what progress I have made so far. So, in order to do the to to get everyone more in the loop with the Evie Weekly, there will be three things that would be very helpful. The first one would be an e-mail to help people who just want to briefly know about what are the updates to stay in the loop.
And then the next thing would be like probably a blog post. So it gives people a more like a bigger picture on like everything happens on the week. And then after that, if people are interested, in a specific snippet, specific topic that get discussed in the Evolve Weekly, then they will straight away jump into the snippet.
Okay, so for the e-mail, mainly how I think about sharing is that it will be like just top three highlights and plus the rest is just a title line only. Later I will show you an example. And then for the blog, it will be like a summary and then each update with a title and with 100 characters of description plus the type and the status of the feature.
And the snippet, when people want to like go into deeper knowing the topic, it would be 480 characters of the summary and then we provide the clips, the VTT, the screenshot and the readable transcript of that. Cool. So let me just show you what the end result first.
So this is one of this is one of the example of the e-mail that I sent out like last last week, the test e-mail. So imagine like the last the last week update, you will get a highlight on, okay, so this is the update and then top three highlights. So I think last week highlight is by Adeline's, one of the things that Adeline's built on Outlook e-mail, one connectors.
And then after that, it will be everything under just also this week. So if people are interested, they can just click on the link and then jump straight in. to the specific item. So then this is the snippet view that I mentioned.
So it will be like the title. And then who are the ones who present it? Is this a use case or is this a feature?
And then is this feature live in free port, live in production?
Nour Shaker (she/her) 5:19
Hello. You okay if I just jump on the call? Yeah, yeah, yeah, I'm silent.
Jecelyn Yeen 5:28
live in production or still work in progress. And then the clip will be already like clip and then the summary is the 480s. And then the screenshot, you can see that this is a screenshot getting from the recording and it also automatically crop out the hits of All the presenters, and then this is the readable transcript. and then go back to the blog post.
This would be the blog post that lists down all the updates in brief. Cool. So how I do this is that.
Maybe I talk about how I imagine this would be done and then how much I have accomplished. So for part one, I can imagine that there will be a web hope or there will be something that once the meeting is finished, the MP4 and the transcript will get an update and then call and web hope and then it will call the EO workflow. and they will produce the first draught of the articles and then in researcher so that a PO or whoever is going to do the edit will be able to get a message from Microsoft team and then just click and then And your researcher, in order to do the review. And then part 2 is when PO click on the link and open the draft, it can edit and then use researcher as the core editor to do some editing and then get it published.
And then publish means that it will push to, it will start the workflow to actually crop the video, like cropping the screenshot, generating the transcript and stuff like that and do and to send the e-mail. So currently all the part that is like on a date can be done and those part that is in currently like in the warning mark, partially it can be done. So let me show you what I can already done now.
Cool. So, in this, in the Evo Weekly, in this researcher, what I have is that I created a Evo Weekly connectors. Yeah, so what I will do is that imagine the video already, video and transcript already update somewhere.
What I can do is I can say that grab. April 10th, VTT is the transcript. and start drafting you all weekly. What this will do is that it will call the connectors.
Hee. It will call the connectors, you see that it will call the API endpoint to try to get the VTT file, the transcript file, and then it will start drafting the Evolve Weekly. So how it starts drafting the Evolve Weekly is that it will utilise the skill that I give it.
Let me show the skill, okay, from here. So I have two skill attached. So one of the skill is the Evo Weekly producer skill.
So I provide like step-by-step information on how an e-mail should be, Evo Weekly should be draft. So the output of this is a JSON file, like a structured format that later on if you are going to draught the e-mail, or draught the post, everything will be using this standard metadata in order to do so. Um...
Okay, so you see that it tried to get the A, let me a pretend. ATT. Let's try again.
So I think just now it tried to search, but it cannot. Let's try again. Yeah, he got it.
Yeah, so it will start drafting. In the meanwhile, I can show you the... Video on the result.
Cool. Okay, never mind, maybe I just show here. So this is the other day that I, the chat that I have earlier.
So one is finished drafting, it should come up with like a JSON file like this. So it's showing that what a list of the feature and then the speaker, the start and end time. that we need to crop the video and then the type of the features, the subtitle and the description and also the screenshot. So I actually asked it to like just pick top three screenshot that's most likely showing a demo screen and then later on we will just crop a video from there.
Sorry, crop a screenshot from there. Okay, but you know, like JSON file is not very editable. So the ideal format is that we have MFE attached to it so that the PO can edit it easier.
So there is also an editor that I have created that can open it up. Let me open it up. So this is the last week data.
So what it will show is that once the JSON finish, it will pass the data into the into the MFE. And then ideally, PO can edit here and then like swipe the feature importance and then try to change because I think one thing probably the AI would get it wrong is that what is the status? Is this life in product or life in pre-port or is what is the type?
So this might be the part that we need to adjusting and some of the wording. So currently, like whatever we edit in the MFE here, right? A nice thing to do is that the MFE will, the result of my edit will go back to the researcher in order to continue the edit, but currently it can't do so.
So you can only call the... You can only like copy back the JSON file to the the JSON content into the EVO with the researcher in order to do the editing. Cool, so once the finish the edit, right, then you can, we can what we can do is that, OK, cool, upload the.
Upload the. Jason. So what this does is that it can also call the, it can also call the connector.
To then upload this, upload this JSON file on the April 10th, go back to the go back to our Microsoft blog. Let me just refresh. So it's still calling, calling it.
So when it's done right, then the next thing is that we can kick off. So ideally, like after it's done, what we can do is that we can just say like, kick off, like start the evil weekly process. In workflow automation.
But currently, I think the team is still doing something on the authentication. So what we need to do is that we need to start it manually here. So we can have a web code.
And then let me wait for the JSON file to upload first. Yeah, okay, cool. Let me refresh.
So the JSON file is here. So that in workflow. Imagine that this can be done automatically.
That first it will try to get the JSON file by the date. So it's the April 10th. And then what it will do is it will try to transform the data. into the format that I want in order to do the cropping video and cropping the transcript.
So you see here in this snipper folder, there is no April 10th transcript yet because I haven't done anything yet. So now let me just start the cropping process. So this process will crop each of the video, probably around 10 of them, and then we create a folder here.
Let's refresh. So you can see right now all the video are here already. And then continue it will also it will also call the cropping the transcript.
So, one is done. You can see the transcript file is here as well. And then the next process is that we need to crop the screenshot as well.
So when it's done, it will also like adding this to the folder. Ohh, wait. Go back.
Hey. Okay, go back here. Snippet.
Okay, it's done. Okay. And then another thing that it's nice to do is that, so now you can see that there is like image screenshot that get cropped.
So one of the thing is that just now at the first step when we generate this image file, right, it's by guess, it's by the transcript. It tried to guess, like, is that a demo slide or is that a screen demo or is that just showing a slide or a talking head? So, a cool thing to have is that we can have an agent builder to do an image classifier.
So, in this image, in this agent builder. What I have attached is that I have attached the open API, I've attached the connector, and then what I can do is that I can chat to it. And then say that classify.
Screen. A pretend screenshot. The need.
So ideally I can hook this agent into the workflow automation. So after like for each of the screenshot that we have, just hook it up and then to make sure that it showed. So Interesting, it cannot call, but I can show you a video that it is working.
Yeah, so it will generate, like, if this type of like web app is the confidence is high or is the confidence is low, so we only will show those screenshot that is like most likely a web app and or because it will be a helpful screenshot for the for the for the readers. Okay. So you can see that part of this already done, part of this is still in progress.
So I think like, is making like evil weekly like a readable format and shareable format is one of the way to make everyone an evil expert and to help everyone stay in the evil loop. Yeah, just by making it easy to reshare. because there's a lot of great stuff in this meeting. It's just that sometimes it's like 5 P.m. here.
It's end of the day and Friday. So yeah, thank you, everyone.