Uploaded March 2016 | Updated November 2026, 3 days ago
I made a hot and moderately super saturated solution of magnesium sulfate and allowed it to cool. I then placed a grain of magnesium sulfate into the solution to observe the crystallisation rate over a period of 5 hours.
1) Weigh out 20 g of magnesium sulfate heptahydrate into a clean empty beaker.
2) Add 15 mL of deionised water.
3) Heat over a Bunsen flame or hot plate until all of the salt has dissolved. It is necessary to place a lid over the beaker to prevent premature and accelerated evaporation, which forms a crystal crust on the surface of the solution.
4) Flash cool the beaker in an ice-bath.
5) Place a crystal grain of magnesium sulfate heptahydrate to investigate crystallisation rate.
As you can see from the video, there was initial growth of the crystal grain. However, the growth seemed to be unstable, and the morphology of the crystal was not preserved. Primary nucleation set in eventually at multiple sites. The set-up could be improved by having a lid to slow down evaporation. It suffices for a simple optimisation run.
I made a hot and moderately super saturated solution of magnesium sulfate and allowed it to cool. I then placed a grain of magnesium sulfate into the solution to observe the crystallisation rate over a period of 5 hours.
1) Weigh out 20 g of magnesium sulfate heptahydrate into a clean empty beaker.
2) Add 15 mL of deionised water.
3) Heat over a Bunsen flame or hot plate until all of the salt has dissolved. It is necessary to place a lid over the beaker to prevent premature and accelerated evaporation, which forms a crystal crust on the surface of the solution.
4) Flash cool the beaker in an ice-bath.
5) Place a crystal grain of magnesium sulfate heptahydrate to investigate crystallisation rate.
As you can see from the video, there was initial growth of the crystal grain. However, the growth seemed to be unstable, and the morphology of the crystal was not preserved. Primary nucleation set in eventually at multiple sites. The set-up could be improved by having a lid to slow down evaporation. It suffices for a simple optimisation run.










![Applied AI Product Management: User Stories for an AI Feature
In this video, youll learn how to write effective user stories for AI-enabled apps—and why you should NOT be building chatbots everywhere.
What Youll Learn:
1. How to write user stories that embed AI into specific workflows
2. A new acceptance criteria format: Scenario-based writing
3. The Smart Breakdown feature example: Auto-generating subtasks from complex tasks
4. How to give LLMs clear instructions while maintaining control over your feature scope
Key Concept: Good AI Design ≠ Chatbots
Sloppy app design puts chatbots everywhere. Good AI design automates specific parts of your workflow seamlessly—users shouldnt even realize theyre interacting with AI.
The Scenario Format:
Given [initial condition]
When [user action]
Then [expected outcome]
This format gives you tighter control over what features the LLM builds, trading off some creativity for precision and reliability.
I built this for our public service community. If you find it useful, share it across the public service, please. And when you build something, tag me on LinkedIn (https://linkedin/in/yongkiat), and Id love to see what you create!
Lets grow together. See you in the videos!
Instagram: / rogue.teacher.sg
User Story:
On the to-do app page, I want to create a Smart Breakdown AI feature using Lovables inbuilt LLM integration with Gemini. Heres the user story:
As a user with a complex or overwhelming task, I want the AI to automatically break it down into smaller, actionable subtasks, so that I know exactly what steps to take and feel less overwhelmed by large projects.
Acceptance Criteria
Given I have created a new task with a title like Plan company retreat,
When I tap the ✨ Smart Breakdown button in the task detail view,
Then the AI should analyze the task and generate 5-10 relevant subtasks within 5 seconds.
Given the AI has generated subtasks,
When the generation is complete,
Then I should see a preview screen showing all suggested subtasks with checkboxes to select which ones to add.
Given I am viewing the suggested subtasks,
When I review the list,
Then each subtask should have a clear, actionable title and be ordered in a logical sequence.
Given the AI suggests subtasks with estimated timelines,
When my original task has a due date,
Then the subtasks should have suggested due dates distributed appropriately before the main deadline.
Given I am satisfied with the AI-generated subtasks,
When I tap Add All or Add Selected,
Then the chosen subtasks should be created under the parent task within 1 second and I should return to the task detail view.
Given I dont like some of the AI suggestions,
When I deselect certain subtasks before adding,
Then only my selected subtasks should be added to the task.
Given I want to make changes to a suggested subtask,
When I tap on a subtask in the preview,
Then I should be able to edit the title, due date, or other details before adding it.
Given the AI fails to generate subtasks (API error, timeout, etc.),
When the error occurs,
Then I should see a friendly error message: Couldnt generate breakdown. Please try again with a retry button. Applied AI Product Management: User Stories for an AI Feature](https://i.ytimg.com/vi/6o62X6G26Gw/mqdefault.jpg)