Intro, demo, and some stats for daigr.am

GPT builders Plugins / Actions builders
,
1

Hi everyone,

I’ve been mostly lurking so far, and wanted to introduce myself as the author of the daigr.am plugin. It’s been over a week since the plugin went live and so I have some stats that may be of interest to others. I used this opportunity to eat my own dog food so to speak and present them as a conversation I just had with ChatGPT to:

  1. figure out how to use KQL (Kusto Query Language) which is what Azure Application Insights uses to query log data. I’ve never used it before today, so i asked for help and then out-of-view, used that to get some data to render.

  2. produce a daigr.am using the resulting output showing the chart creation request failure and success counts over the past 7 days

  3. adjust the visualization to better show what is intended

Here’s the entire conversation: https://chat.openai.com/share/5307da71-4837-4812-a224-7b37bae8fd3d

As you can see, my initial plug-in manifest could use a lot of improvement. Right now it takes the model an average of three requests for each successful chart creation. I think the main issue is that the request format is over-generalized to try and handle many chart types when it should be more specific to each chart type. I think creating specific request formats with meaningful field names (like groups, series, lines, etc) will help. I’ve held off making any changes to the manifest for now until the “unable to uninstall unverified plugin” issue is resolved.

I’ve had, and seen others have, a lot of success using my plugin along with other plugins to render charts with the data they produce: BlockAtlas, WebPilot, KeymateAI Search, and many others.

Hope this data and demo helps… Lots more to come!

-Ray

1 Like
2

On the chat you shared, it shows at least 66% of the requests to the service result in a 500 error. Is that correct? I wonder if that’s affecting the users experience?

I am also curious about how the backend services you use to render these graphs before showing them to the user.

3

There is an example of an error in the conversation, where the first attempt to make the chart fails. The model just tries again using the explanation I provide in the response. I’ve since changed the response to be a “400” response code since it’s an error in the request body format, not a service bug. I don’t think most users are affected except having to wait for the re-submission by the model.

The backend is entirely new code written by me for this plugin. It’s designed to work similar to D3.js with the main differences being it’s typescript and it can render markup strings to an output stream as well as rehydrate from DOM when loaded in the browser. This all runs as an Azure Functions app with Azure API Management to route requests.

1 Like
4

Wow! that’s a great project

1 Like
5

Thanks @finadon, and welcome to the community!

6

Welcome to the forums, cool Plugin! Thanks for sharing your experience. Hope it helps budding pluggers to build new things.

1 Like
7

Here’s an update to the daigr.am chart creation requests stats. Several improvements were made to handle malformed requests on the backend without updating our ai-plugin.json manifest and there is a marked improvement in the failure/success ratio for the most recent week.

image
image1920×1331 70.2 KB

On June 7 a bug was introduced causing the service to crash after failed creation requests, and you can see a major downturn in the stats, but that problem was quickly resolved.

Still holding off submitting an updated manifest due to the ongoing “unable to uninstall unverified plugin” issue. Once that is fixed, I expect the number of failed creation attempts to reduce significantly.

Here’s the conversation I used to create this area chart: https://chat.openai.com/share/68818966-b904-4ef1-9870-c76c702700c2

1 Like
8

Why can’t Diagr.am render this JSON?
Code Interpreter has no problems with it.
Did I do something wrong?

{
“version”: “1.0”,
“model”: {
“nestingLevels”: [
{
“name”: “Years”,
“labels”: [“2023”, “2022”, “2021”, “2020”, “2019”, “2018”, “2017”, “2016”, “2015”, “2014”],
“aggregator”: “first”
},
{
“name”: “Countries”,
“labels”: [“India”, “Germany”, “USA”, “China”],
“aggregator”: “first”
}
],
“data”: [
[
[1428627663, 1417173173, 1407563842, 1396387127, 1383112050, 1369003306, 1354195680, 1338636340, 1322866505, 1307246509],
[83294633, 83369843, 83408554, 83328988, 83148141, 82896696, 82624374, 82331423, 82073226, 81858825],
[339996563, 338289857, 336997624, 335942003, 334319671, 332140037, 329791231, 327210198, 324607776, 322033964],
[1425671352, 1425887337, 1425893465, 1424929781, 1421864031, 1417069468, 1410275957, 1401889681, 1393715448, 1385189668]
]
]
},
“view”: {
“title”: “Einwohnerzahlen (2014-2023)”,
“type”: “Line Chart”,
“scale”: “linear”,
“notes”: “Einwohnerzahlen von Indien, Deutschland, USA und China für die letzten 10 Jahre.”
}
}