Generative AI Application Retrofit - Startup Briefing (S1:E3)

Movie in LinkedIn (S1:E3)

In today's post, I'm excited to share how we're transforming the startup experience in our application with ChatGPT. We've introduced a dynamic Startup Briefing feature, offering a unique blend of technology and user-centric design. This feature is more than just an informative display; it's an interactive guide to kickstart the user's day.

Imagine starting your application and being greeted by a personalized briefing. This is made possible by presenting ChatGPT with Key Performance Indicators (KPIs) and allowing it to generate a concise, relevant summary tailored to the user's objectives. It's akin to having a personal assistant who highlights critical items and actions needed to jumpstart your day.

Startup Briefing as drawn from the KPIs

At application startup, a function collects the KPIs, and ChatGPT crafts a readable summary of essential actions. This process, which adds approximately 30 seconds to the startup time, is a small investment for the significant value it brings. It's designed to run at the beginning of a work session or after prolonged inactivity, ensuring that the insights are timely and pertinent.

We're thrilled about the potential of this feature to enhance the user interaction with our application. It's not just about providing information; it's about enhancing productivity and decision-making right from the start.


Chat Popup with Startup Briefing

To construct the prompt for this analysis, I selected key KPIs that I had previously established on the application’s Dashboard screen, focusing on those that offer the most insightful data for our user’s objectives. Each KPI included a title, a description, and a count. For the two most critical KPIs, I also incorporated a few example records. The resulting prompt, which included this data, was structured so that the first paragraph defined the problem and instructions for ChatGPT, with the remainder providing the full context of what to analyze.

I am a writer working on submitting my stories to publishers. I need to be attentive to alerts or deadlines and the status of my submissions, also unpaid expenses or sales. Evaluate the following input to generate an analysis with less than 200 words. For each analysis item, show the name in bold format and then the count within parenthesis. Sequence these items in the order of most urgent to least important. Format the analysis as a bullet list. Details should be included with an item if provided. Format the details with that item as an additional indent level using a '+' for the bullet character.

Alerts Incomplete and Past Due 46 Incomplete and Past Due| Details include: aaa - Follow-up - 5/10/2020Andromeda Spaceways Inflight Magazine | Reading... - Deadline - 4/1/2020
Log Entries for Samuel Grayson - - 10/23/2019
|Alerts Due Today 0 Okay.
Submissions Overdue 54 Review overdue submissions to close out and re-submit.| Details include: My UI | Bewildering Stories | 11/17/2017 - 11/17/2017
New Story Name | A.K. Press/ A.K.A. Books... - 12/19/2019
A Frozen Fire | 1913 Press | 10/7/2017 - 10/7/2017
|Submissions Pending 57 Okay.
Stories Available to be Sent 0 Okay: All available inventory has been sent out.
Stories Pending a Response 35 Okay.
Events for Today 0 Okay. No events scheduled for today.
Events for this Week 0 Okay. No events scheduled for this week.
Sales with Payment Due 29 Check on these 29 outstanding balances due to you with a total of: $10.00.
Expenses Not Paid 3 Check on these 3 outstanding balances you owe for a total of: $0.00

Input Generated for ChatGPT

For generating the Startup Briefing, we've taken a dynamic approach to analyzing pre-defined KPIs. While these KPIs already offer some level of analysis, leveraging ChatGPT has allowed us to transform this data into more concise, reader-friendly insights. Drawing from the Dashboard page, we've moved beyond the lengthy and mechanically-generated output to a format that resonates more with our audience. By guiding ChatGPT to present information in bullet points, coupled with a brief narrative, we've created an output style that's not only popular among our users but also enhances clarity and engagement.

- Alerts Incomplete and Past Due (46)
  + Details include: aaa - Follow-up - 5/10/2020
  + Andromeda Spaceways Inflight Magazine | Reading... - Deadline - 4/1/2020
  + Log Entries for Samuel Grayson - 10/23/2019
- Submissions Overdue (54)
  + Review overdue submissions to close out and re-submit.
  + Details include: My UI | Bewildering Stories | 11/17/2017 - 11/17/2017
  + New Story Name | A.K. Press/ A.K.A. Books... - 12/19/2019
  + A Frozen Fire | 1913 Press | 10/7/2017 - 10/7/2017
- Sales with Payment Due (29)
  + Check on these 29 outstanding balances due to you with a total of: $10.00.
- Expenses Not Paid (3)
  + Check on these 3 outstanding balances you owe for a total of: $0.00.
- Submissions Pending (57)
- Stories Pending a Response (35)
- Stories Available to be Sent (0)
- Events for Today (0)
- Events for this Week (0)

The most urgent items in the analysis are the Alerts Incomplete and Past Due, as they require immediate attention and follow-up. These include a specific alert with the name "aaa" that needs a follow-up by 5/10/2020, as well as an incomplete reading for Andromeda Spaceways Inflight Magazine with a deadline of 4/1/2020. The Submissions Overdue category is also important, as it involves reviewing overdue submissions and resubmitting them. There are 54 submissions in this category, including My UI at Bewildering Stories with a submission date of 11/17/2017. The Sales with Payment Due category should be addressed next, with 29 outstanding balances totaling $10.00. The Expenses Not Paid category is less urgent, with only 3 outstanding balances totaling $0.00. The remaining categories, including Submissions Pending, Stories Pending a Response, Stories Available to be Sent, and Events for Today/This Week, do not require immediate attention.

Please go to the Workflow tab and use the Dashboard function for more details.

Output Generated by ChatGPT

Behind the Scenes

We've enhanced our startup sequence with a new function, defining the instructional part of the prompt on line 40. It then generates the KPI information on lines 41 to 50. Then, on line 52, we integrate the KPI information along with sample records into the prompt. The magic happens in line 65, where we make the call to the ChatGPT API.


Function “Generate NLP Startup”

Lessons Learned

The instructions in our prompts needed several tweaks to truly resonate with our users. To make the experience more relatable and practical, we included a few sample detail records. This approach goes beyond the standard, mechanically-generated Dashboard screen, offering users a meaningful starting point for their day's work.

In this journey, we discovered that the ChatGPT-4 model was a good fit, prioritizing quality over performance. In the future, we anticipate that new and improved models will further enhance our output. I believe it is important that we recognize this is a dynamic process that will change over time. To ensure we're always on the right track, we've implemented a mechanism to gather user feedback on each prompt. This allows us to monitor and adapt to changes effectively as we evolve with the technology.

Stay tuned for my next update, where I'll dive into how we're leveraging Natural Language Processing (NLP) for searching and modifying data in the system, and performing basic automation operations. Don't forget to FOLLOW me here for my latest insights and developments, see (S1:E4)!

1 Like