📄 Stream Notes
Created by Ajay at EntryLevel Get 10% off an EntryLevel course using coupon code SHADOWPM10
View the stream video here: https://youtu.be/Q6B29P5OB_s
Key Learnings
- Use customer interviews to validate a problem however if you want to validate a solution you should use live data prototypes. You shouldn’t be asking users if they would use or might like something. Those are subjective questions. Use objective metrics like usage, time on site etc.
- Building a Live Data Prototype → a quick, valuable test helps us build conviction to continue on this pathway
- There are a few ways to ask users if they like your product
- Net Promoter Score
- Sean Ellis Test
- Rate out of 5 stars
Subscribe
I write about various things and update my collection of models & questions. If you want to hear about that, you can subscribe:
PRD: Personalised Learning
This is the PRD we developed live on stream - it is completed unedited since we streamed! This is what we made within the 2 hour period 😀
1 Context / Vision
Users tend to be quite different in their feedback on specific content and missions. Some say it’s too easy and some say it’s too hard.
We want to explore how use AI to personalise the learning journey
2 Goals / Metrics
Primary Metric: End Course NPS
GOAL: {55, 70}
Guardrail:
- Maximum Churn Rate
- On Mission Feedback Average
- Conversion Rate (Sales → How much of our web traffic buys the course)
3 Core Problems & User Stories
User Stories
- User complains the content is too difficult
- User complains the content is hard to read or understand
- Users forget to do the modules
- Users think modules take too much time
- Users don’t fully understand previous module when they start the new module
Learning Designer Stories
- LD does not know how to actual feedback when they are polar opposites.
- LD has a lot of ideas for personalisation but no way to implement it
- LD want to engage learners in the zone of proximal development
- LD want custom scaffolding for each learner
- LD want to engage learners based on their individuals interest
4 Explore the Current Build
- Lack of personalisation on Ajay AI
- Content is static
- Quizzes are also static
5 Research Solutions
- Audio Summaries by Google
- Generate a podcast based on the content using AI
- Users can interrupt and ask a question
- Synthesis School
- AI Voice helping them answer questions without giving them the answer - “Hints”
- GUIDED PROJECTS
6 Solution Scope
My product is always in a SHIPPABLE STATE.
What’s in Scope
- Create audio experiences for learners based on the content to allow them to listen to modules (maybe as they wash dishes)
- Interest Based Tailoring of the Content
- AI Generated Quizzes
- AI Tailored DIFFICULTY of content
- AI based reordering of modules & generation of new modules
Not in the Scope
- AI Guided Projects
7 Solution Breakdown
1 Interest Based Personalisation
This screen will appear right after onboarding steps. It will be optional.
For each module pick a random topic from the selected topics.
Option 1: Feed the entire module into AI & ask it to change examples to match that topic
Option 2: Create ‘example blocks’ so it’s explicit which parts of the course is examples vs just learning.
2 Instant Podcasts
- User selects a mission and the mission that is auto-selected will be the “next mission”
- The AI generates a transcript for a conversation that is in the style of a podcast
- Get an AI to format the script → SUGGESTION: Text → JSON (Speaker A: dawdawd, Speaker B: dadwa)
- Elevenlabs for the voice → generate as a stream
- Play Audio
- If user asks a question either as voice or text. Pause the narration
- Feed Current Script or left over script & question into AI
- Regenerate AUDIO
- Whenever the AI Is loading display the “generating response” text
Mobile Version additional features
- Play whilst closed
- Voice Only interruptions
- Focus Mode (get rid of distracting dashboard features)
3 Quiz Based Content Difficulty Modulation
- Feed modules into AI and generate a quiz
- Generate correct & wrong answers in a table
- Store User’s Correct & Incorrect Answers
User the user’s correct & incorrect answers to determine a difficult coefficient. And feed entire module into AI to modulate. Also use this survey at the end of each mission to modulate difficult. Use the input from the PREVIOUS mission & previous mission generated content.
Based on
{Previous mission content}
the user said it was {Too easy, Just Right, Too Difficult}
Please adjust this content according to the feedback:
{New Mission}
At the end of each MODULE (Multiple Missions) create a custom revision module. By inserting all of the content in the previous module and which quiz questions they got right and wrong. Focus on content for the ones they got wrong!
8 Implementation Breakdown
Phase 1: Interest Based Personalisation (1-2 weeks)
- Have the onboarding form
- Each time the user loads a mission, it feeds the ENTIRE mission → AI with a random pick of their interests
Phase 2: Difficulty Modulation
- How did you find this mission question
- Edit the next mission
Phase 3: Difficulty Modulation
- AI Generated Quizzes
- Using the Quizzes to Modulate the next content
- Custom Revision Missions at the end of each module
Phase 4: Instant Podcasts
- No Interupts
- Every podcast can be generated ONCE
Phase 5: Instant Podcasts w/ Live Changes
- Users can interupt and get their questions answered
Created by Ajay at EntryLevel Get 10% off an EntryLevel course using coupon code SHADOWPM10
This is a live PRD we created on a livestream which you can view here: https://youtu.be/Q6B29P5OB_s
Subscribe
If you subscribe - I will send you my best work. A collection of 100+ mental models, questions I like to ask and more!