schoolportal.ai

Currently I have developed the solution for Pymble Ladies' College am now moving to commercialise and deploy the solution to other schools.

A new entity will be incorporated, founded and led by me and have full ownership of the IP and will commercialise this exciting solution.

In a world where AI creates new content, how can teachers assess student work to check mastery? Are they assessing the student or the tool they are using? To help address this we are embedding teacher-controlled AI tutors within a school portal and learning management systems. 

• The teacher will limit how the AI can respond
• the AI will adapt to the student's demographics, interests and abilities
• the AI will use RAG to know the subject material and task the student is required to do and provide appropriate sanctioned assistance to the student.

Furthermore, the teacher will be able to view a history of how the students have engaged with the AI and coach and support the students with greater insight.

By having the Assistant conversations hosted in Microsoft Azure, the data is secure and private.

The AI assistant data is hosted within the Microsoft Azure tenant of each school, providing the highest level of data governance and security for personal data.

This allows us to safely provide AI assistants to teachers of K-8 learners, and where appropriate, the learners themselves.

Rather than outsourcing the access to AI to external companies, and therefore losing all insights into how the students are engaging with the assistants, the AI and conversational history is within the School's digital safe bubble and allows safe and supported guided use of AI tools appropriate for each age group and demographic.

The AI will also provide an insight into how the students are using the assistant - providing an indication of how much support, how off topic, or how passive the learner was in any interaction. This will provide insights to the teacher and position them as guide and mentor through the AI learning journey.

The Portal and AI assistants are also language and device agnostic and will respond in the most appropriate language and comprehension level for each learner based on the user RAG model that is foundational to the GPT.

By using APIs to surface the best and most appropriate AI model, we can ensure the cost is per use, not per user, and will make the price of the solution dramatically more cost effective for the school.

We have already built the language agnostic, data agnostic, device agnostic, branding agnostic portal and completed the first phase of delivering an embedded AI assistant that can be deployed to any subgroup within a school. This has been in production at one of Australia's premier schools with over 4000 users.

As the lead of this initiative, I am well placed to deliver the ideal solution for K-12 schools. I have been a teacher or worked with schools since 1991. I ran a software business for 14 years developing student administration and portal solutions, before selling the business to a multinational in 2013 and returning to teaching and being responsible for IT innovation at Pymble Ladies' College - Australia's leading Independent Girl's school with over 2400 students.

With our current Portal solution and embedded AI, for the past 5 years, we have worked closely with all members of our school community and delivered constant improvements based on the feedback of over 4000 people.

We have been using the Portal since 2021 with our parent community. We are now rolling out this portal to staff and students in May 2024 and have piloted the use of our embedded AI tool with students in both the Junior and Secondary schools along with staff.

This tool is now well accepted across our school, and we act as our own pilot program for this solution.

1. Allowing the teachers to control how the assistant operates ensures staff ownership of AI and that it augments them, not replace them.
2. Allowing the end user to set communication preferences increases adoption.
3. Feeding the AI assistant RAG based information about the user ensures personalised targeted appropriate support during the assessment task.
4. Feeding the AI assistant RAG based information about the subject area and associated textbooks helps the AI to provide more trustable responses.
5. Limiting the conversation to the specific assessment/ learning task, under the guidance of the teacher, personalised by the student ensures targeted meaningful learning support.
6. Keeping the conversations private within the School's Azure tenant keeps students safe.
7. Providing teachers with insights into how the students are engaging with the AI allows the teacher to coach and mentor the student rather than being blind to how students are using AI in a learning task, including a sentiment analysis of how much support the student needed.
8. Clearly showing the students how the assistant can be used helps students understand their responsibilities and frees them up to grow with curiosity and with AI assistance without fear of being labelled a cheat.
9. Embedding AI within Portal that aggregates over 15 different solutions into one device agnostic solution.
10. Being Language and branding agnostic ensure the solution will be appropriate for any community and diverse user base.
11. Allowing our AI assistants to be embedded into external tools through LTI can surface the assistant wherever the user needs it.

• AI Conversation Storage: hosted in Microsoft Azure
• AI Model: Agnostic, but currently favouring OpenAI GPT4-turbo - mediated through Microsoft Azure.
• RAG: Textbook RAG in partnership with Copyright Australia
• Teacher curated Content RAG through OpenAI Fetch-file or LangChain/ Pinecone RAG.
• User RAG based on Student Admin System data.
• LangChain and OpenAI function calling retrieves data from various RAG solutions such as GroundX, Pinecone, SchoolJoy.
• Portal is developed in React, TypeScript, Python, CSS, JS, Ionic and relies on some licensed third-party tools for some UI component.
• Enterprise Service Bus store is MongoDB hosted on Atlas.
• REST APIs delivered through FastAPI and Azure API services.
• Data syncing done through Dagster.
• We have a single codebase and build to Web (PWA), iOS, Android using Capacitor.

1. We have deployed the Portal Solution to over 4000 parents for over 2 years.
2. We have been piloting the portal with 500 staff and 2500 students for 6 months. 
3. We have been piloting the embedded AI assistants with staff and students for 6 months.

You can view some of the Portal documentation here. Note - this documentation is a couple of years old and is only about the portal, and NOT all the AI we have developed and embedded into the solution. I share this link only to demonstrate it is beyond the concept phase and the technology works.

• The AI tutor responds in the primary language of the user, based both on the base demographic information we have stored, and also their own preference.
• We also pass the age/ reading age of the user to ensure the language of the response is not just in the right language, but also at the right level.
• We provide each user with the ability to set the style/ tone of the language to add some additional personalisation.
• Teachers are able to provide guardrails on how the assistant is to respond to ensure it stays on task.
• Microsoft provide a 'content safety' layer that can ensure the response from both AI and users meet safety criteria
• The AI can be instructed at a system level to challenge bias, provide a diversity of opinion and to always lead conversations with kindness. Concept proof of this was through establishing a debate coach and also a 'values as behaviour' coach that aligned actions with our school values.
• Teachers gain insights into how each user is engaging with the GPT and gets a sentiment analysis of the support level, depth of autonomy and focus demonstrated by the student to put them into a position of coaching each student in ethical appropriate use of the AI

Looking at the screenshots below, you can see the difference in how the same assistant automatically responds to the same question for an English speaking adult, a young English-speaking student and a young Spanish-speaking student:

English speaking adult

Young English speaking student

Young Spanish speaking student

Five people are working on this project.

We are ready to pilot with a new school now. 

We are currently in negotiations with a school with over 3000 students to be our second school using our AI assistants and embed them into their learning management system.

When we first created the Portal, we ensured it was Internationalised. We also ensured that we used an Enterprise Service Bus model so that we could integrate with any system on the edge, and still ensure our Portal solution would work.

One of the major blocks to wider adoption of high-quality GPT4 or compatible grade AI tools is they are charged on a per user basis. For OpenAI licensing, 1000 users would cost $25,000 a month.

By providing our AI assistants that leverage the API GPT4 or equivalent models, we are able to deliver GPT4 level quality on a per use, rather than per user basis.

This brings the cost of high-quality AI access down to under 10 cents per in-depth conversation and makes the solution much more affordable for all students and staff to access.

By switching the model to lower quality or free solutions like Llama or Mistral, these costs can be further reduced.

We are seeking investment to help ramp up our commercialisation of schoolportal.ai and the Personalised AI tutor for every student.

We are seeking both financial investment and channel partners to embed our solution within their products and communities.

We are seeking legal assistance to ensure the successful IP transfer and future independence of the company.