Applying prompts to GenAI/LLMs

Imagine buying complex furniture, but the instructions are a chaotic pile of sticky notes. This is often the daily reality for software engineers. Whether you are "vibe coding" a new feature or trying to connect two different web services, you often find yourself digging through messy, undocumented code just to understand how things talk to each other. The OpenAPI Specification (OAS) is the "Universal Instruction Manual" that fixes this. Why This Matters for Modern Engineering I n a world of AI agents like Gemini, ChatGPT, and Claude , having a machine-readable "map" of your code is a superpower. For Humans: It provides a clear map of what a service does and what it needs. For AI Agents: Tools like Cursor or Windsurf can use an OpenAPI spec to understand your project’s "intent" (the vibe) without getting lost in the syntax of your legacy files. For Students: It allows you to take an old Python/Flask project and instantly generate interactiv...

From Google Scholar to Project Ideas or Using AI to Map the Future of Your Research Generated as well by Google Gemini Have you ever looked at a researcher’s Google Scholar profile and felt overwhelmed? Between the long lists of citations and dense technical titles, "connecting the dots" of a decade-long career is a massive cognitive lift. Whether you are a student hunting for a dissertation topic or a professional scouting for a collaborator, understanding the trajectory of research is harder than reading the papers themselves. In my latest experiment, I used Google Gemini to see if it could bridge this gap. I gave it a specific challenge: Analyze my own research profile and design 10 compelling project ideas for a final-year Computer Science student. The Prompt: Turning Data into Direction The secret to a good AI output is giving it a clear "anchor." By providing a URL to a live dataset (my Scholar profile), I bypassed the need to copy-paste thousands of words....

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 GenAI Productivity: Ideas to project proposal 1 GenAI Productivity: Ideas to project proposal 2 One of the most significant cognitive hurdles for students is the transition from a high-level area of interest to a rigorously defined project specification. This "blank page" problem often leads to poorly scoped projects or a lack of alignment with current academic literature. Building on my previous experiments in prompt-mediated productivity, this post examines a structured workflow for using Large Language Models (LLMs)—specifically GPT-4o—to scaffold the development of a BSc Software Engineering project proposal. The Methodology: Multi-Stage Prompt Scaffolding Rather than requesting a single output, the workflow utilizes a recursive prompting strategy. This mirrors the iterative refinement pr...

OpenAI has just announced canvas ( https://openai.com/index/introducing-canvas/  ) with ChatGPT as a beta in  ChatGPT Plus . Canvas is an extra interface along with the standard prompt interface which we have got used to.  In an earlier post, I started discussing using Canvas for  reports,  but what about coding? People have been using generative AIs for coding, including code generation from prompts, but what does it do when we use canvas. So let's play. The new interface for canvas does support coding (as does ChatGPT) - could it be a programmer's friend/assistant?  In the example below ChatGPT was asked via an initial prompt to produce a pseudo-code for the start of a murder-mystery style game, but to also use canvas.  What it produces is a form pseudo-code in the style of Python; not a great surprise there, as ChatGPT is not bad at generating Python code. So, can it convert the pseudo-code to something other than Python? The two examples below sho...

OpenAI has just announced canvas ( https://openai.com/index/introducing-canvas/  ) with ChatGPT as a beta in  ChatGPT Plus . Canvas is an extra interface along with the standard text interface for ChatGPT prompts and responses. It feels more like using a featured text editor.  At the moment OpenAI are rolling out canvas to ChatGPT Plus and Team users, then   Enterprise and Edu users will get access next week; with the plan to  make canvas available to all ChatGPT Free users after the beta stage. So to have a bit of a play I put an old document through it (selected the ChatGPT 4o with canvas option from the  ChatGPT Plus drop-down menu), a skeing it to use that document as a starting point for a report. One of the difference is both windows are editable for prompts. So using the lefthand side you can put in a prompt as normal  in ChatGPT but in the right-hand wind you highlight the text and ask  ChatGPT  do action that bit .  You can...

Image created using DALL-E - love the bad spelling I want to explore using Generative AI to explore and compare websites. So I used using Google Gemini because of it its ability to work easily with websites.I choose UK three political party websites purely to compare like with like. Prompt 1: Analyse the following webpages website 1 https://www.conservatives.com/ , website 2 https://www.libdems.org.uk/  and website 3 https://labour.org.uk/ from a web user's perspective. For each website produce a report containing 2 tables. The first Table list issues with the site; for each issues provide at least three examples and then provide a list of potential solutions. Table 2 strengths of the site with each strength provide at least three examples. Add in any commentary Results It produced two tables per website and provided a summary comparing them at the end. Direct political statements were not produce. Now trying out personas and testing the website the politics side not filtered out ...