#Onions and Prompts

I recently came across a genuinely useful idea on Tom's Guide about using an “onion prompt” with AI to organise your schedule when you’re overwhelmed. If you haven’t seen it, I’d strongly recommend reading the original article — it explains the thinking behind the method and the psychological principles that make it effective:

https://www.tomsguide.com/ai/i-use-the-onion-prompt-with-chatgpt-when-im-buried-in-tasks-it-cuts-through-clutter-in-seconds

How I’ve adapted it

I’m using Google Gemini rather than ChatGPT, and I used my task list from Google Keep instead of viewing my desktop. The tools are little different (but not too much), but the central ideas are the same: strip away the layers hiding your real priorities and let AI “peel back” your to-do list until only what truly matters remains.

To make it easier, I copied my Google Keep list into a Google Doc and pasted it into the prompt.

---

Prompt 1: Prioritising with the “Onion” Method

Here’s the version I’m using (slightly adapted from the Tom’s Guide example):

I feel buried under competing tasks. Here is my to-do list:
"<paste your to-do list>"

Peel back the layers and categorise items into:
• Core (essential progress)
• Important (schedule soon)
• Surface noise (quick admin)
• Remove (close or ignore)

Then identify the top 3 priorities for the next 90 minutes.

Did it work?

Yes, it worked better than I expected.

• It factored in the realistic 90-minute window I had at the end of the day.

• It grouped related tasks into coherent activities.

• It highlighted where I was at risk of “prep panic” before meetings.

• It clearly identified what could safely be ignored.

Prompt 2: Turning Priorities into Time Blocks

I then followed up with a second prompt:

Taking into account that activities such as “XXXXX Slides & 3x Activities (Due 5/3)” take 1 hour each, identify time blocks to complete all tasks across the week.

This is where it became even more powerful.

It:

• Split the week into structured blocks.

• Themed days (e.g., “Meeting Marathon” and “Deep Thinking”).

• Matched the type of work to appropriate times (lighter admin vs. cognitively demanding tasks).

That alignment between task type and time of day made the plan feel far more realistic and sustainable.

The original article on Tom's Guide  https://www.tomsguide.com/ai/i-use-the-onion-prompt-with-chatgpt-when-im-buried-in-tasks-it-cuts-through-clutter-in-seconds goes into more detail about the psychological theory behind the method — particularly how cognitive overload hides priorities under layers of low-value activity.

If you often feel buried under competing demands, it’s worth trying.

Give the approach a go — and do read the original post https://www.tomsguide.com/ai/i-use-the-onion-prompt-with-chatgpt-when-im-buried-in-tasks-it-cuts-through-clutter-in-seconds for the deeper reasoning behind it

All opinions in this blog are the Author's and should not in any way be seen as reflecting the views of any organisation the Author has any association with. Twitter @scottturneruon

Reverse engineering a specification from a solution using GenAI: Part 1

Imagine buying complex furniture, but the instructions are a chaotic pile of sticky notes. That is exactly how programmers feel when trying to connect different web services (like weather apps or payment processors). They usually have to dig through messy documentation just to make two programs talk to each other.

The OpenAPI Specification (OAS) fixes this by acting as a universal, standardised "Instruction Manual."

Why It Matters

OpenAPI acts as a perfect bridge between humans and machines:

• For Humans: It provides a clear map of what a service does, what information it needs, and what it returns.

• For Computers: Because the rules are strict and predictable, software can "read" the manual and automatically connect to the service.

What Can You Do With It?

With an OpenAPI blueprint, developers can plug into tools that do the heavy lifting for them:

• Create Visual Guides: Turn complex code into sleek, interactive websites where users can test the service with the click of a button.

• Write Code Automatically: Instantly generate the "glue code" needed for apps or servers, saving hours of manual typing.

• Automate Testing: Let tools read the blueprint to automatically double-check that the software works exactly as promised.

The Big Question: Can AI Work Backwards?

Here is where it gets interesting. If we have existing software but no instruction manual, can we use Generative AI to "reverse-engineer" one just by looking at the code? Doing so would instantly unlock all the automated benefits mentioned above for older projects.

In this post, we are putting this to the test: Can ChatGPT take a zipped folder of existing code and successfully work backwards to write the OpenAPI blueprint?

Step 1: A zipped file for an example Python Flask based simple payroll system (and it is simple): https://github.com/scottturnercanterbury/musical-meme.git

Step 2: Load the zipped file into ChatGPT and the following prompt was used "Unpack this zip file into individual files. Produce an OpenAPI specification based on these files starting with app.py"

It went through the process of unzipping, following the struct , ignoring virtual environment properties, etc.

Produce a YAML filr

Step 3: Using the prompt "Converted the OpenAPI specification in YAML to JSON and show"

{

  "openapi": "3.1.0",

  "info": {

    "title": "Musical Meme Payroll Management System",

    "version": "1.0.0",

    "description": "A simple Flask-based payroll management application for managing employee records with salary tracking. This service is primarily HTML-rendering (server-side templates) and uses form submissions rather than JSON APIs."

  },

  "servers": [

    {

      "url": "http://localhost:5000"

    }

  ],

  "tags": [

    {

      "name": "Employees",

      "description": "Employee payroll records"

    }

  ],

  "paths": {

    "/": {

      "get": {

        "tags": ["Employees"],

        "summary": "List employees",

        "description": "Renders a page listing all employee salary records.",

        "responses": {

          "200": {

            "description": "HTML page containing employee list",

            "content": {

              "text/html": {

                "schema": {

                  "type": "string"

                }

              }

            }

          }

        }

      }

    },

    "/employees/add": {

      "get": {

        "tags": ["Employees"],

        "summary": "Show add employee form",

        "responses": {

          "200": {

            "description": "HTML form to add an employee",

            "content": {

              "text/html": {

                "schema": {

                  "type": "string"

                }

              }

            }

          }

        }

      },

figure 1. An extract of the returned specification.

Not a bad start! Let's check if it is valid 

Using https://swagger.io/  and setting up a new account. We can load it in and then check if it works against the specification (hint:it did)

Figure 2 Swagger testing the specification

figure 3: The schemas produced 

As an experiment, can the process be done in reverse ie. go from specification to code: see next post.

All opinions in this blog are the Author's and should not in any way be seen as reflecting the views of any organisation the Author has any association with. Twitter @scottturneruon

Ideas to project proposal ideas from Google Scholar - the video

GenAI Productivity: Ideas to project proposal ideas from Google Scholar

Generated as well by Google Gemini

In previous blog posts, I have looked at generating project ideas for project ideas. These can be used with many different GenAI platforms. As an idea/challenge, I want to come up with a way for students to generate ideas for projects based on knowing who they would like to work with and that person's Google Scholar profile.

Here is the catch: often, it can be difficult to access Google Scholar to do this. One solution  is to use Google Gemini, a Google product, to access another Google product. Going to apply this to my own profile - ego rides again. It went and did much when I click on it and deep research, it did a full research report.

Prompt used:  "Using this as a starting point https://scholar.google.com/citations?user=ghQedZAAAAAJ&hl=en from the research here provide 10 project ideas suitable for a final year Computer Science student project with this supervisor. For each provide title, 100 word summary, possible outcomes"

Results and unforeseen consequences

It ended up producing a quite extensive report.

Yes it produced ten project ideas and an extract is shown below (it even promoted me to Professor, which I don't mind)

"

2. Proposed Final Year Computer Science Projects

Each of the following project ideas is designed to leverage Professor Turner's research strengths and align with current advancements in Computer Science.

2.1 Project 1: Adaptive Multi-Robot Path Planning in Dynamic Environments using Deep Reinforcement Learning

This project proposes developing a deep reinforcement learning (DRL) framework for multi-robot path planning in highly dynamic and unpredictable environments. Building upon Professor Turner's foundational work in probabilistic multi-robot path planning , this endeavor will explore how DRL agents can learn optimal navigation strategies, collision avoidance, and task allocation in real-time. The system would adapt to unforeseen obstacles and changes in robot capabilities, aiming for robust and efficient coordination in complex scenarios.  

Possible Outcomes:

• A simulated multi-robot environment with dynamic obstacles.

• A DRL model trained for decentralized path planning and collision avoidance.

• Performance evaluation comparing DRL-based planning with traditional algorithms (e.g., A*, DFS as in ) on metrics like path length, collision rate, and computational efficiency.  

• Analysis of the DRL agent's adaptability to novel dynamic scenarios.

Underlying Trends, Thematic Links, and Implications: This project directly extends Professor Turner's existing research on "Probabilistic Multi Robot Path Planning in Dynamic Environments: A Comparison between A* and DFS". His prior work already addresses the complexities of dynamic environments and the necessity for efficient pathfinding.  

A significant progression from traditional search algorithms like A* and DFS, which Professor Turner's earlier work explored , involves embracing adaptive AI techniques. While A* and DFS are effective for planning in known or semi-known environments, they can be less flexible when confronted with entirely unforeseen dynamic changes or highly complex, non-deterministic conditions without extensive pre-programming or continuous re-planning. Current trends in AI and robotics emphasize the development of systems that are "smarter, more adaptive, and autonomous". Reinforcement Learning, a field experiencing exponential growth , offers a paradigm where agents learn optimal policies through direct interaction and trial-and-error within their environment. This inherent learning capability makes DRL particularly well-suited for dynamic and uncertain conditions. Therefore, integrating DRL into multi-robot path planning moves the research beyond merely comparing fixed search algorithms to developing truly intelligent, self-learning robotic systems capable of emergent behavior in complex, real-world-like scenarios, marking a substantial advancement in robotic autonomy. The concept of "Physical AI," where robots are trained in virtual environments before deployment, is also highly relevant here.  

Furthermore, multi-robot systems, as explored in Professor Turner's publications , inherently face scalability challenges as the number of robots and the complexity of the environment increase. Traditional centralized planning methods can become computationally prohibitive. DRL, especially when coupled with decentralized learning approaches, offers a promising avenue to overcome this. The capacity for DRL agents to "make independent decisions" aligns with the need for decentralized control in large robot swarms. This project could explore how individual DRL agents learn to cooperate implicitly without a central planner, thereby addressing a key scalability bottleneck in multi-robot systems and paving the way for more robust and fault-tolerant robotic deployments.  

2.2 Project 2: AI-Driven Predictive Maintenance for Software-Defined Networks (SDN) in 5G Environments

This project aims to develop an AI-driven system for predictive maintenance and performance optimization in Software-Defined Wide Area Networks (SDN-WANs), particularly relevant for emerging 5G infrastructure. Leveraging Professor Turner's expertise in SDN routing and AI , the system will analyze network telemetry data (e.g., traffic patterns, QoS metrics, device health) to predict potential failures, congestion, or security vulnerabilities before they impact service. This represents a shift from reactive network management to proactive, intelligent control.  

Possible Outcomes:

• A simulated SDN-WAN environment (e.g., Mininet, as used in ) integrated with a data collection module.  

• Machine learning models (e.g., LSTM, anomaly detection algorithms) trained on synthetic or real-world network data for predicting network issues.

• A prototype SDN controller module demonstrating AI-driven alerts or automated reconfigurations for predictive maintenance.

• Evaluation of prediction accuracy, false positive rates, and the system's impact on network uptime and QoS.

Underlying Trends, Thematic Links, and Implications: This project directly builds on Professor Turner's highly cited work, "Routing algorithm optimization for software defined network WAN" , which focuses on improving SDN-WAN performance. It also connects to his broader interest in "AI".  

Professor Turner's SDN routing work primarily focuses on optimizing routes based on current network conditions to enhance Quality of Service (QoS) and reduce delay. However, modern networks are inherently dynamic, and issues such as congestion and security vulnerabilities can arise unpredictably. Current trends in the SDN industry explicitly highlight the "Integration of AI for predictive maintenance and performance optimization". This signifies a fundamental shift from merely reacting to network problems to proactively preventing them. By integrating AI for  

predictive maintenance, this project elevates the scope of network management from optimizing the current state to anticipating future states. This capability is crucial for the demanding requirements of 5G and IoT applications, where network reliability and efficiency are paramount.

While SDN offers significant flexibility and enhanced network management capabilities, the separation of the control plane from the data plane can introduce "complex security concerns". The increasing integration of AI for network automation and predictive maintenance creates new potential attack vectors if not properly secured. A critical consideration, therefore, is how an AI-driven predictive maintenance system, while enhancing efficiency, must also be robust against adversarial attacks or data poisoning, which could lead to incorrect predictions or malicious reconfigurations. The project could explore how the AI models themselves contribute to, or are vulnerable to, security issues, and how to embed "enhanced security solutions" directly into the AI-driven SDN management plane.  

•    
•    
•    

2.10 Project 10: Robotics Simulation Platform for Experiential Learning of AI and Multi-Agent Systems

This project proposes developing an interactive robotics simulation platform designed for experiential learning of AI algorithms, particularly focusing on multi-agent systems and problem-solving. Inspired by Professor Turner's interest in "Robotics within the teaching of problem-Solving" and the trend of "Physical AI" where robots train in virtual environments , the platform would allow students to design, test, and visualize AI agents controlling virtual robots in various problem-solving scenarios (e.g., maze navigation, resource gathering, cooperative tasks).  

Possible Outcomes:

• A 3D simulation environment (e.g., using Unity, Gazebo, or a custom engine) capable of rendering multiple robots and dynamic obstacles.

• API for students to implement and integrate their own AI algorithms (e.g., pathfinding, reinforcement learning, genetic algorithms).

• Pre-built problem-solving scenarios (e.g., multi-robot search and rescue, collaborative assembly).

• Visualization tools for debugging AI behavior and analyzing simulation metrics.

• User interface for scenario creation and parameter tuning.

Underlying Trends, Thematic Links, and Implications: This project directly builds on Professor Turner's "Teaching and Learning" area, specifically "Robotics within the teaching of problem-Solving" and "Problem Solving and Creativity for Undergraduate Engineers".  

Professor Turner's work emphasizes the process of problem-solving and creativity. Traditional Computer Science education often involves theoretical study and coding exercises. The progression here is that a robotics simulation platform offers a powerful "experiential learning" environment, allowing students to directly observe the real-time consequences of their AI algorithms in a dynamic, visual context. This bridges the gap between abstract theory and practical application, reinforcing understanding of concepts like path planning , multi-agent coordination, and the nuances of AI algorithm design. This approach makes the learning process more engaging and effective, aligning with the "Physical AI" trend where robots train in virtual environments.  

Real-world robotics experiments are often expensive, time-consuming, and prone to hardware failures. The "Physical AI" trend highlights the value of virtual environments for training and experimentation. A critical consideration is that a robust simulation platform can provide students with access to complex, large-scale, and even dangerous multi-agent scenarios that would be impossible or impractical to replicate in a physical lab. This scalability allows for a much broader range of educational experiences, from simple maze solving to simulating complex factory automation or disaster response scenarios, all without the physical constraints. This effectively democratizes access to advanced robotics and AI experimentation, fostering innovation among a wider student body.  

"

The unexpected outcome is that it did a good analysis of my research from the most cited papers, which I wasn't expecting, including a table (see below) relating the papers to current computing trends and potential directions.

1.2 Overview of Professor Scott J. Turner's Research Landscape

1.3 Key Observations from Professor Turner's Profile and Current Trends

1.4 Table 1: Alignment of Professor Turner's Expertise with Current CS Trends

Professor Turner's Primary Research Area	Relevant Publications/Keywords	Current CS Trends (2024-2025)	Potential Synergy/Project Direction
Artificial Intelligence (AI) and Robotics	AI, robots, genetic algorithms, Probabilistic Multi Robot Path Planning	AI in robotics (autonomous systems, physical AI, cobots), Reinforcement Learning (RL) for autonomous systems	Adaptive robot control, multi-agent systems, AI-driven simulation for robotics
Networking and Communication Systems	SDN, WebRTC, V2V Communication, Routing algorithm optimization, Reinforcement learning-based routing	SDN with AI/5G/IoT, WebRTC with AI/AR/VR, V2V for road safety/traffic management, RL for networking	Intelligent network management, real-time multimedia communication enhancements, smart vehicular networks
Biomedical and Health Applications	Optical glucose detector, Modelling chronic pain, Evoked potentials	AI-powered diagnostics, personalized medicine, predictive analytics for patient care, remote patient monitoring	AI for medical data analysis, smart health monitoring systems, advanced diagnostic tools
Teaching and Learning / Problem Solving	Teaching and Learning, problem-solving, creativity, Robots in problem-solving	AI agents, intelligent automation, data foundation for AI in education	AI-driven educational tools, interactive learning platforms for complex problem-solving

But also an analysis of themes and future directions of research

3.1 Synthesis of Project Themes and Interdisciplinary Connections

3.2 Recommendations for Further Exploration

So I got something out of it as well the ideas generation - this is very cool. It allows you to copy to Google Docs and so you can make it available that way. 

All opinions in this blog are the Author's and should not in any way be seen as reflecting the views of any organisation the Author has any association with. Twitter @scottturneruon

GenAI Productivity: Ideas to project proposal 3

Produced using Google Gemini

In two previous posts I looked at using Generative AI to start producing  project ideas and refining one to be the start of the proposal, Previous blog posts

1. GenAI Productivity: Ideas to project proposal 1
2. GenAI Productivity: Ideas to project proposal 2

This post is really a slightly different variation on those. Playing with adding a few things in. The example can be repurposed for non-computing student but the example came from a discussion with a BSc Software Engineering student, thinking about project ideas

Example starting prompt in ChatGPT4o

"Produce 5 project ideas suitable for a BSc Software Engineering students for the following topic: "Ambient Assisted Living" The project must suit someone with a strong interest in programming. For each idea produce a 100 word description; justification in 200 words why this is a worthwhile project;

a set of goals; sources of potential data; physical resources needed"

It does produce 5 projects - personally, they do look interesting. Now  selecting a project to expand and add references

"Taking project 4. , Add in appropriate references throughout the whole project idea. Expand the justification to 500 words and the Project Description to 200 words. At the end, add a reference list. Harvard referencing throughout. Referencing should be in every section. Show the whole project idea"

Final output

"

Project Title:

Voice-Controlled Personal Assistant for Dementia Patients

---

Project Description (200 words):

This project involves designing and developing a voice-controlled personal assistant (VCPA) specifically tailored for individuals living with dementia. The assistant will use natural language processing (NLP) and speech recognition to offer intuitive, speech-based interaction that accommodates cognitive impairments. Features will include answering basic orientation questions (e.g., “What day is it?”, “Where am I?”), issuing routine reminders (e.g., medication, hydration, appointments), and engaging users with conversational prompts for companionship.

Unlike general-purpose devices like Amazon Alexa or Google Assistant, this system will be customized to address the specific needs of dementia users, such as simplified dialogue, memory aids, and emotionally supportive interactions (Bainbridge et al., 2021). The assistant will avoid complex instructions and provide gentle redirection when confusion is detected.

Built using open-source tools (e.g., Mozilla DeepSpeech, Rasa, or Vosk), the system will run on a low-cost device such as a Raspberry Pi and feature a caregiver dashboard for configuring responses and monitoring interaction patterns. The project will focus on low cognitive load design, customisable content, and offline or private data handling to enhance privacy and usability (Topol, 2019).

By developing this assistant, the project will demonstrate practical programming, AI implementation, and human-centered design tailored to assistive technology applications (Blackman et al., 2021).

---

Justification (500 words):

Dementia, affecting over 55 million people globally (WHO, 2023), presents challenges in memory, orientation, and daily living. As cognitive decline progresses, individuals may forget essential routines, become disoriented, or experience increased social isolation. While caregivers provide vital support, constant supervision is often unfeasible, especially given rising healthcare demands and staffing shortages (Prince et al., 2015). Assistive technologies like voice-controlled personal assistants (VCPAs) offer a promising solution to support independence and reduce caregiver burden.

Commercial solutions such as Amazon Alexa or Google Assistant are not optimised for people with dementia (Kerssens et al., 2015). They typically assume a high degree of digital literacy and short-term memory, and may cause frustration when misunderstood. These devices are also not tailored to emotional or contextual support. Research by Moyle et al. (2018) shows that when digital companions are personalised and accessible, they can improve quality of life and reduce anxiety in dementia patients.

This project addresses a clear gap in existing systems by focusing on cognitive accessibility, emotional reassurance, and task support through naturalistic speech interaction. Simplified command structures, context-based reminders, and a limited response set will be used to reduce cognitive overload, as recommended by Hung et al. (2020). The assistant will use Natural Language Processing (NLP) and dialogue management techniques to provide consistent responses and adapt to individual needs.

From a software engineering perspective, the project encompasses a range of core skills: real-time system design, speech interface development, dialogue management, and user-centered design. NLP models can be implemented using open-source tools such as Rasa or Snips, allowing the student to work with intent classification, entity recognition, and response generation (Vinyals & Le, 2015). Offline voice recognition using Vosk or DeepSpeech supports privacy-preserving development, which is especially important when handling sensitive health-related information (Topol, 2019).

This project is valuable not only technically, but ethically. Developing a technology that genuinely assists vulnerable individuals aligns with calls for “ethically aligned design” in AI (IEEE, 2019). It also provides students with an opportunity to engage in inclusive and person-centered technology creation, reflecting a broader shift towards responsible innovation in the software engineering field (van Wynsberghe, 2016).

Given the growing prevalence of dementia, especially in aging societies, this project is timely and socially impactful. It offers scope for interdisciplinary learning, practical AI development, and contributes meaningfully to the goals of Ambient Assisted Living (AAL). The final product can serve as a prototype for further research or commercial development.

---

Goals:

1. Develop a simplified NLP interface for use by dementia patients.

2. Implement a voice-controlled reminder and question-answer system.

3. Design a user-friendly caregiver interface for content management.

4. Enable offline speech recognition for privacy.

5. Evaluate usability using dementia-friendly design heuristics.

---

Potential Data Sources:

• DailyDialog dataset for natural conversation modelling (Li et al., 2017)

• WHO Dementia Behavioural Guidelines (WHO, 2023)

• Existing caregiver support scripts from Alzheimer’s Society (Alzheimer’s Society, 2022)

• Persona-based testing scenarios from academic dementia studies (Hung et al., 2020)

---

Physical Resources Needed:

• Raspberry Pi 4 (or equivalent SBC)

• USB Microphone and Speaker

• Internet connection for updates (optional if offline models used)

• PC for training dialogue models and dashboard development

• External display (for caregiver dashboard)

---

References:

Alzheimer’s Society. (2022). Dementia guide for carers and family. [online] Available at: https://www.alzheimers.org.uk [Accessed 10 Jun. 2025].

Bainbridge, E., Potts, H.W.W., & Wyatt, J.C. (2021). Digital health for the cognitively impaired: The challenge of inclusivity. Journal of Medical Internet Research, 23(2), e16850.

Blackman, T., et al. (2021). Ambient Assisted Living technologies to support aging in place: Evidence and implications. Ageing & Society, 41(8), 1792–1810.

Hung, L., et al. (2020). The use of intelligent assistive technologies for people with dementia: A scoping review. Journal of Biomedical Informatics, 108, 103484.

IEEE. (2019). Ethically Aligned Design: A Vision for Prioritizing Human Well-being with Autonomous and Intelligent Systems. IEEE Standards Association.

Kerssens, C., et al. (2015). Personalized technology to support older adults with and without cognitive impairment living at home. American Journal of Alzheimer’s Disease & Other Dementias, 30(1), 85–97.

Li, Y., Su, H., Shen, X., Li, W., Cao, Z., & Niu, S. (2017). DailyDialog: A manually labelled multi-turn dialogue dataset. Proceedings of IJCNLP 2017, 986–995.

Moyle, W., et al. (2018). Exploring the effect of companion robots on emotional expression in older adults with dementia. Journal of Gerontological Nursing, 44(7), 35–41.

Prince, M., et al. (2015). World Alzheimer Report 2015: The Global Impact of Dementia. Alzheimer’s Disease International.

Topol, E. (2019). Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again. Basic Books.

van Wynsberghe, A. (2016). Ethics for robots: How to design a moral machine. Science and Engineering Ethics, 22(2), 415–431.

Vinyals, O. & Le, Q. (2015). A neural conversational model. arXiv preprint arXiv:1506.05869.

WHO. (2023). Dementia. World Health Organization. [online] Available at: https://www.who.int/news-room/fact-sheets/detail/dementia [Accessed 10 Jun. 2025]."

Next stages: go through it and apply the human bit:

• Are references real or even good? 
• You must read the ones you include.
• Could go back in and get it to find sources for physical devices
• Is it 'doable'?
• Rewrite for the particular context.
• Think of it as a starting point, not the end; refining it will make a better bit of work.

All opinions in this blog are the Author's and should not in any way be seen as reflecting the views of any organisation the Author has any association with. Twitter @scottturneruon