This blog post reflects on the development of my Digital Artefact over Semester 2, 2024. What started as an exploration into brand development has evolved into a personal data visualisation project aimed at helping users understand the narrative their data tells. Inspired by projects like Spotify Wrapped, my goal is to explore how people can engage with their data in meaningful and insightful ways, as well as aggregating it all together to highlight its power when combined.
Initial Concept: My initial project idea, BuildingBrands, was framed around creating a digital platform for brand development. However, after reflecting on my own interests in personal data and automation, I decided to pivot. This shift led me to reframe my project to focus on the research question: What insights can be gained from intercepting the personal data collected by my technology and apps?
This pivot allowed me to explore how disconnected data points could be aggregated into a coherent narrative, not only about who I am but also for predicting future trends. My inspiration drew from studies such as Nicol et al. (2022), which emphasise the risks and complexities of understanding digital traces. As well as AI-specific work from Agrawal, et al., (2022), which forces us to think about AI and its connection to prediction, referring to AI as Prediction Machines. A recent manuscript from Scott Cunningham & Van Pham deeply inspired this investigation. The article looks at claims that "ChatGPT Can Predict the Future When it Tells Stories Set in the Future About the Past" (Van & Cunningham, 2024).
A Place Where All Data Comes Together
Midjourney: "dramatic digital illustration AI servers, database, networking"
At the start of my project, I faced the challenge of integrating various personal data streams into one cohesive platform, with the goal of creating a personalised summary similar to Spotify Wrapped. I initially turned to ChatGPT for data processing, but its limitations in handling large amounts of data forced me to seek alternatives. This led me to experiment with Glide, a no-code platform that integrates GPT-4. By repurposing Glide for my data aggregation needs, I applied the concept of exaptation (Garud et al., 2016), using the platform way beyond its original design. This shift allowed me to start generating weekly summaries from my data, which offered early insights into patterns of app usage, productivity, and spending habits.
Establishing Additional Data Streams
As I progressed, I realised that my initial dataset was not comprehensive enough to provide a full picture of my behaviour. Exploring the adjacent possibilities to this stage of my project let to me expanding my data collection to include new sources such as GPS data from Life360, daily financial transaction summaries, and even images from my camera roll. By integrating these additional data streams, it allowed me to uncover new insights from my data. The more data sources I integrated, the more intricate the narratives became. For example, I could infer my work-from-home habits based on app usage during business hours, demonstrating the potential of combining various data points to create a detailed profile.
Automating for Project Sustainability
Midjourney: "AI servers, database, networking https://s.mj.run/fzV3zdErdwc --ar 7:3"
As my dataset expanded, it became clear that manual data processing was not sustainable in the long term. This realisation prompted me to explore automation tools such as iTools, which enabled bulk data exports, and Google Sheets for more streamlined processing. By automating data collection, I aligned my project with the concept of antifragility, where systems grow stronger when exposed to challenges (Taleb, 2012). By setting up some type of automation, the project would become much more sustainable but also be more resilient, ensuring that it could continue functioning even with inconsistent human input. Due to restrictions on IOS, I would have required certain developer privileges and the app still wouldn't pass approvals from the IOS app store due to it basically being an app that transmits all data to cloud. In recent updates, Apple has cracked down on privacy, with a suite of new requirements to allow access to most parts of the IOS. As a result, I really struggled to develop something myself to collect all the data on the phone to the same detail as i was collecting manually, and couldn't find anything which existed that i could use instead.
Creating a Real-Time Database For Connecting To Other Apps
Midjourney "dramatic digital illustration AI servers, database, blue ethernet networking cables https://s.mj.run/fzV3zdErdwc --ar 7:3"
During class discussions about Web3.0, I reflected on how personal data in the future might interface directly with platforms rather than being stored in proprietary databases. I read an article by Lans, (2012) that discussed the creation of personal data stores, which aligned with the above Web3.0 concept, where production platforms merely interface with the personal data stores to bring a tailored experience to each user.
My peer reviews highlighted the need for making my project helpful to others, which meant I had to move my data off my computer and into a dynamic database. The Glide database I was using worked well, but it was limited in accessibility from other platforms. (Knight, et al., 2017) suggests using Google Forms and Sheets as a database for collaborative metadata creation in digital projects due to its low cost, and ease-of-use. While unconventional, it functioned as a modular, cell-based system that allowed easy imports and exports of CSV files, fitting perfectly with my text-based data.
After some research and with the help of ChatGPT, I generated a Google App Script that converted the live Google Sheets file into a URL formatted as JSON. This allowed me to link the sheet to various tools, creating a dynamic, real-time system where changes made in the sheet updated instantly through the URL. The ability to fetch and update data dynamically was a breakthrough, as it created a truly accessible and scalable knowledge base. This shift aligned with my ongoing exploration of exaptation, where I've been pushing platforms beyond their intended purposes, allowing me to achieve interoperability without building an expensive, hosted database.
(Nano Tutorials, 2016)
This process taught me the value of pushing existing tools to their limits. By using Google Sheets in a way it wasn’t designed for, I achieved something practical and adaptable, which allowed me to use my data in more places, making it more valuable moving forward (Díaz & Contell, 2018). I’m considering how this dynamic database can facilitate broader applications, not just for my project but potentially for other users who could benefit from this real-time data connection (Li & Clifton, 1995) & (Xia, et al., 2002). This decision underscores the adjacent possibilities that arise when we creatively reimagine existing tools and systems.
Trialling Various Media Formats To Interact With My Data
Midjourney "dramatic digital illustration AI servers, database, blue ethernet networking cables https://s.mj.run/fzV3zdErdwc --ar 7:3"
Initial peer feedback suggested expanding the utility of my project beyond personal reflection, encouraging me to explore how this data-driven assistant could offer proactive advice to users. This feedback led me to explore the idea of exploring a recommendation engine that could personalise suggestions based on accumulated data, whether through daily insights, trend predictions, or behavioural guidance. This adjacent possibility of my raw data being collected would allow the project to evolve from a personal tool into something that could benefit a wider audience, making data more meaningful and actionable in everyday life.
I trialled creating an AI assistant that could interact with this live data through my above real-time database. While tools like Intercom offered an engaging conversational interface, it had limitations with handling personal data, often returning generic responses due to its built-in safety features.
Exploring OpenAI revealed both challenges and opportunities, as it occasionally failed to read all context data due to character limitations, but could deliver useful insights when it did.
(Flanagan, 2024)
The potential for AI to generate personalised suggestions based on a user’s trends—like recommending productivity breaks or budgeting tips—introduced new avenues for my project’s utility.
models generated with ChatGPT v4o
Using AI to Build a User Profile from Data
Midjourney "dramatic digital illustration database creating user profiles, demographics https://s.mj.run/x1xl76KB1IY --ar 7:3"
From the outset of my project, my goal was to explore how personal data collected from phones could be used to build a detailed user profile, often without the user's awareness of the extent of the data being gathered. Initially, I tried using Intercom to extract demographic data from my dataset, but its built-in safety features limited my analysis. This led me to pivot to OpenAI’s development playground, where I tested the AI by inputting a month’s worth of my personal data to see if it could form an accurate profile. I was pretty impressed as despite the data being anonymised, the AI made surprisingly precise assumptions about my profession, income range, and lifestyle habits, demonstrating the ease with which data can be used for marketing insights.
| Demographic Factor | Profile Information |
|---|---|
| Speculated Age | Mid to Late 20s |
| Gender | Male |
| Ethnicity/Race | Anglo-Australian or European descent |
| Location | Australia (NSW) |
| Education Level | Bachelor's Degree |
| Income Level | Estimated $70,000 - $100,000 AUD |
| Occupation | Media Production, Marketing, Project Manager, Small Business Owner |
| Marital Status | Likely Single |
| Household Size | 1 (Solo) |
| Home Ownership | Likely Rents |
| Language | English |
| Nationality | Australian |
| Internet Access | High-speed access, frequent usage |
| Employment Status | Employed Full-time or multiple Casual/Part-Time Roles |
| Social Class | Middle Class |
| Transportation | Access to both private vehicle and public transportation |
| Financial Stability | Moderate Savings, Investment in Small Business, Emergency Fund |
| Living Standards | High; access to all basic amenities, living in an urban metro area |
This experiment illuminated the power of AI to connect seemingly unrelated data points, a process that aligns with the concept of exaptation. As a result of this milestone, I started looking into custom GPT models within OpenAI directly to try and establish a GPT which other's could use to achieve something similar through AI data interpretation.
| Attribute | Details |
|---|---|
| Technology Use | Heavy user of multiple devices (Apple products, apps like Life360, fitness trackers, automation tools). |
| Health & Fitness | Active; frequent walking and occasional running. Focus on weight loss and fitness improvement. |
| Diet | Low-calorie diet (around 1,000 calories per day). |
| Time Management | Structured and disciplined schedule, managing professional, academic, and personal tasks efficiently. |
| Professional Tools | Uses AI and automation tools like Zapier, and services such as Shopify, Reckon, and POS systems. |
| Creative Interests | Engages in media production, marketing, design (working on candle fragrance, digital artefact projects). |
| Academic Focus | Active university student, pursuing a degree in communication and media, with a focus on digital and social media. |
| Business Mindset | Entrepreneurial, involved in managing multiple ventures, including small businesses (candles and media). |
| Data-Driven | Regularly collects and analyses personal data, including fitness, app usage, and project performance. |
| Cultural Engagement | Connected to community events, such as the Cultural Spectacular, and involved in event management and promotion. |
| Creativity & Innovation | Involved in creative projects, including digital artefacts, advertising campaigns, and experimental work. |
| Environmental Awareness | Potential consideration for environmental and social impact (e.g., product design, local event engagement). |
| Mental Health Awareness | Data hints at a focus on work-life balance and well-being through structured routines and fitness efforts. |
| Music Interests | Engages in music streaming; top songs suggest a taste for modern, alternative R&B and pop genres. |
| Technological Savvy | Comfortable with integrating advanced technologies such as AI, automation tools, and smart devices into daily life. |
The process also revealed adjacent possibilities, where the limits of existing tools are expanded to create new uses. Through this, I’ve gained a deeper understanding of how AI-driven personalisation could (and is) become a significant tool for marketers and developers.
It has reshaped my perspective on the power of data aggregation, while raising important questions about privacy, data ownership, and the ethical implications of using personal information in the digital age.
Using My Learnings of the Semester to Create a Custom GPT
Midjourney "dramatic digital illustration AI servers, database, blue ethernet networking cables https://s.mj.run/x1xl76KB1IY --ar 7:3"
Over the semester, I gathered data from my blog posts, peer reviews, and personal information from September, why not share that info so others could do something similar. My objective was to develop a tool where users could input their data and receive detailed assumptions about their lives, similar to how marketers create profiles for targeted campaigns. To achieve this, I compiled all the data and input it into OpenAI’s Playground, which generated raw instructions to build the custom GPT. I then tested the model with various data inputs, observing how accurately it made assumptions based on the collected data.
Instructions for Custom OpenAI GPT Data Assumption Model
Objective:
The purpose of this GPT model is to take user-provided datasets and return direct assumptions or insights based on patterns observed within the data. The model should strictly adhere to analyzing the data and avoid conversational responses. If assumptions cannot be made definitively, the model will indicate that the data is inconclusive.
Key Points for Assumptions:
- Pattern Recognition: Detect consistent or repeating trends in the data to inform assumptions. For instance:
- Weight Data: If a user provides weekly weight data that shows a steady decrease over time, assume they might be on a diet or involved in an exercise regimen.
- Resting Heart Rate Data: If daily resting heart rate data shows irregular spikes, assume that on those specific days, the user may have experienced stress, physical activity, or illness.
- Contextual Understanding: Make data-driven assumptions based on the nature of the dataset:
- Screen Time Data: If a user logs a high number of hours in social media apps, assume they might be involved in a social media-related activity or exhibit patterns of habitual phone use.
- Follower Count Data: If the dataset shows a sudden spike in followers, assume a viral or popular post may have contributed to the growth, depending on the timing.
- Data Source Integration: When data includes multiple streams, such as app usage combined with physical activity, combine these for broader inferences:
- GPS and App Usage: Frequent use of work-related apps in specific geographical locations can suggest remote work or traveling for business purposes.
- Financial Data: If financial transactions indicate regular spending at fitness centers or healthy food outlets, it supports an assumption of a fitness-focused lifestyle.
- Inconclusive Data: If the data does not clearly indicate a trend, or if the variations are too minor to infer specific behavior, state: "The data provided is inconclusive for making detailed assumptions."
Example Responses:
Example 1: Weight Data (Weekly Weigh-ins)
Data: [Week 1: 80kg, Week 2: 79.5kg, Week 3: 78.9kg]
Response: "The data shows a consistent reduction in weight across multiple weeks. This suggests the individual may be following a diet or exercise plan aimed at weight loss."
Example 2: Resting Heart Rate (Daily)
Data: [Day 1: 60 bpm, Day 2: 62 bpm, Day 3: 72 bpm, Day 4: 60 bpm]
Response: "There is a noticeable spike in resting heart rate on Day 3. This could indicate a temporary increase in physical exertion or stress on that day."
Example 3: Screen Time (Daily)
Data: [Day 1: 7 hours, Day 2: 6.8 hours, Day 3: 7.5 hours]
Response: "The user exhibits high screen time over the provided days. This suggests regular or habitual use of their device, possibly for work or entertainment purposes."
Assumption Rules:
- Time-Based Data: For time-sensitive data such as daily logs (e.g., sleep hours, screen time, or app usage), assume habits or routines if trends persist over multiple days.
- Location-Based Data: If GPS data shows frequent visits to specific locations, such as gyms or workplaces, assumptions about professional or personal habits can be made.
- Unrelated Data Streams: If multiple data streams are disconnected (e.g., financial data and health metrics), avoid combining them unless clear interrelations are evident.
- Demographic Insights: If data indirectly hints at demographic information (e.g., usage patterns or spending behaviors typically associated with specific age groups), assumptions should cautiously consider general societal norms (e.g., younger users being more active on social media).
Data Sensitivity and Ethical Boundaries:
When data could imply sensitive or private matters (e.g., financial transactions or health data), refrain from making assumptions beyond what is directly observable from the dataset. If patterns or assumptions risk breaching personal privacy, state: "No assumption can be made without further context to avoid incorrect conclusions."
Flexibility in Assumptions:
The model should be flexible enough to handle various types of data, such as:
- Activity Logs: Identify patterns in physical activity (e.g., steps per day, workouts).
- Social Media Metrics: Make assumptions about engagement trends based on follower growth or post interactions.
- Health Metrics: Infer lifestyle choices based on patterns in sleep, calorie intake, or exercise frequency.
HTML embed created using ChatGPT.
When I applied this approach, the AI’s ability to connect unique data points and provide insightful feedback was impressive. The model could infer lifestyle habits, working patterns, and demographic information, all from anonymised data. This aligned with the concept of exaptation from Garud et al. (2016), as I repurposed OpenAI’s original intent to generate highly personalised profiles. From this, I learned how AI tools can be adapted for media creation, demonstrating their utility beyond traditional marketing. The ability of the AI to construct detailed user profiles highlights to me that such a tool has some good potential for uses in personalisation and marketing.
Vala Afshar
Reflecting on these insights, I decided to focus my next steps on refining the training data and improving the model's ability to interpret inputs across different contexts, including data which wasn't included within my collected data set, such as follower counts, for example.
Original GPT, with very formal conversation response.
I also created rules around how assumptions are created. For example, to ensure more helpful trends are recognised, "avoid combining disconnected data sources unless clear interrelations are evident.", more importantly, if data indirectly hints towards demographic information, assumptions should be made based on existing knowledge around that demographic. For example, we can assume younger users are more active on social media. These guidelines embed into the training data of the GPT aim to guide the data analysis process, encouraging ChatGPT to call on its existing knowledge to fill the blanks in the data (Van & Cunningham, 2024).
Final version of GPT, with more conversational interface, mimicking the Spotify Wrapped brand voice. Changed response voice to be more engaging and allow people with less data knowledge find useful.
Creating a custom GPT marks a key milestone in project sustainability, allowing this project to be accessed by other people and also allowing me to resume working on this from time to time. As I move into a marketing role, I am constantly exploring how AI can support more personalised experiences for customers.
Wrapping Up
Midjourney "dramatic digital illustration AI servers, networking, guy at cliff looking at servers river data stream https://s.mj.run/jrrgezM7RZ8https://s.mj.run/svQwA9tlzcY --ar 7:3"
This project was a valuable learning experience that highlighted the power of data aggregation and personalisation. By pivoting from my original brand development concept to a personal data visualisation experiment, I was able to explore how disconnected data points can form very coherent narratives about a user's behaviour, habits, and even future trends. Through challenges like managing large datasets and trying to automate processes, I discovered how existing tools could be adapted to create new possibilities through exaption. Elements like generating predictions, identifying trends and creating profiles about the data were encouraging and helps answer my research question, that some rich data can in-fact be gained from from intercepting the personal data collected by my technology and apps?
While this project is not intended to be a finished product or a pathway to success, it has provided me with a stronger understanding of the potential and limitations of personal data currently. This also highlighted the importance of approaching data responsibly, especially in terms of privacy and ethical implications. This also allowed me to visualise why large online corporations place so much value on user data. This final exploration has illuminated new pathways, particularly within AI-driven personalisation and the art of data-centric storytelling. It has left me with insights I am eager to carry forward, setting the stage for impactful, data-informed strategies in my future marketing work. With fresh clarity, I’m already exploring platforms that harness personal data to shape uniquely tailored customer experiences, ready to bridge this knowledge into the next chapter.
References
Afshar, V., 2023. How to achieve hyper-personalization using generative AI platforms. [Online]
Available at: https://www.zdnet.com/article/how-to-achieve-hyper-personalization-using-generative-ai-platforms/
[Accessed 24 10 2024].
Agrawal, A., Gans, J. & Goldfarb, A., 2023. Prediction Machines: The Simple Economics of Artificial Intelligence. 2 ed. Beaverton, United States: Harvard Business Review Press.
Apple Inc, 2024. Membership Details. [Online]
Available at: https://developer.apple.com/programs/whats-included/?ref=uow.jamesmctaggart.com
[Accessed 24 10 2024].
Caruana, V., 2024. How AI-powered personalization is transforming the user and customer experience. [Online]
Available at: https://www.algolia.com/blog/ai/how-ai-powered-personalization-is-transforming-the-user-and-customer-experience/
[Accessed 24 10 2024].
Díaz, O. & Contell, J., 2018. Educating Users to Formulate Questions in Q&A Platforms: A Scaffold for Google Sheets. International Conference on Advanced Information Systems Engineering, p. 154–169.
Flanagan, K., 2024. ChatGPT Is LAZY... Here's Why. [Online]
Available at: https://www.youtube.com/watch?v=Me-FoJwFDtA
[Accessed 24 10 2024].
Garud, R., Gehman, J. & Giuliani, A., 2016. Technological Exaptation: A Narrative Approach. Industrial and Corporate Change, 25(1), p. 149–166.
Glide, 2024. Add AI to your apps in minutes As simple as editing a spreadsheet Try now. [Online]
Available at: https://www.glideapps.com/ai
[Accessed 26 10 2024].
Hart, J., 2023. YOU are NOT prepared for Apple’s New Privacy Requirements for iOS App Developers: Comprehensive Guide. [Online]
Available at: https://medium.com/@emt.joshhart/a-comprehensive-guide-to-apples-new-privacy-manifest-requirements-for-ios-app-developers-d004dc47ad35
[Accessed 26 10 2024].
Intercom, 2024. The new age of customer service is AI-first. [Online]
Available at: https://www.intercom.com/
[Accessed 26 10 2024].
Knight, C., Rodrigues, E. & Ciota, R., 2017. Collaborating for metadata creation on digital projects: using Google Forms and Sheets. LIBRARY HI TECH NEWS, pp. 20-23.
Lans, R. F. v. d., 2012. Personal Data Store. Data Virtualization for Business Intelligence Systems, Issue 1, pp. 27-57.
Life360, n.d. How Does Life360 Work?. [Online]
Available at: https://www.life360.com/learn/how-does-life360-work/
[Accessed 25 10 2024].
Li, W.-S. & Clifton, C., 1995. Dynamic Integration in Multidatabase Systems (1). Journal of Database Management, 7(1), pp. 28-29.
Mahdawi, A., 2024. What is going on with ChatGPT?. [Online]
Available at: https://www.theguardian.com/commentisfree/2024/jan/12/chatgpt-problems-lazy
[Accessed 25 10 2024].
McTaggart, J., 2024. BCM302 Data Tracking Project - JSON Strings via Google Sheet. [Online]
Available at: https://script.google.com/macros/s/AKfycbx8e--1j5ZpRPgKL84jTGmh5KBigUXfLdwGFAB1LK824bLQA9jfM4khO1fgk4R-Kyk/exec
[Accessed 22 10 2024].
McTaggart, J., 2024. BuildingBrands for BCM302. [Online]
Available at: https://uow.jamesmctaggart.com/buildingbrands-for-bcm302/
[Accessed 26 10 2024].
McTaggart, J., 2024. Data Storyteller: Friendly, lively data insights for accessible analysis. [Online]
Available at: https://chatgpt.com/g/g-ZkKwaEQiM-data-storyteller
[Accessed 25 10 2024].
McTaggart, J., 2024. Data Tracking Project - BCM302. [Online]
Available at: https://bcm302-jamesmctaggart.glide.page/dl/d8fc49
[Accessed 24 10 2024].
McTaggart, J., 2024. Exploring Board Game Boundaries, AI-Driven Myth-Making with ChatGPT, and OpenAI’s Role in Data-Driven Narrative Construction. [Online]
Available at: https://uow.jamesmctaggart.com/peer-review-bcm302/
[Accessed 25 10 2024].
Nano Tutorials, 2016. How To Export JSON From Google Sheets. [Online]
Available at: https://www.youtube.com/watch?v=HnFwl-8aTmQ
[Accessed 26 10 2024].
Nicol, E. et al., 2022. Revealing Cumulative Risks in Online Personal Information. Proceedings of the ACM on Human-Computer Interaction, Volume 6, pp. Pages 1 - 25.
OpenAI, 2024. OpenAI Playground. [Online]
Available at: https://platform.openai.com/playground/chat?models=gpt-4o
[Accessed 24 10 2024].
Sgobba, N., 2024. Understanding Technology Exaptation: A Catalyst for Innovation & Creative Approaches to the Unconventional. [Online]
Available at: https://medium.com/@nicksgobba/understanding-technology-exaptation-a-catalyst-for-innovation-creative-approaches-to-the-4f53efa6de2b
[Accessed 26 10 2024].
Van, P. H. & Cunningham, S., 2024. ChatGPT Can Predict the Future when it Tells Stories Set in the Future About the Past.
Wallen, J., 2024. Convert a Google Spreadsheet to JSON-Formatted Text. [Online]
Available at: https://thenewstack.io/how-to-convert-google-spreadsheet-to-json-formatted-text/
[Accessed 26 10 2024].
Xia, Y., Stinner, R. & Chu, P.-C., 2002. Database integration with the Web for biologists to share data and information. EJB Electronic Journal of Biotechnology, 5(2).
