From Brainstorm to Build
I first learned about the potential of closed GPTs through a professional development session led by Sam Mormando of Garnet Valley School District. That session inspired a new line of thinking: What if we trained an AI exclusively on our district documents, board policies, and Pennsylvania school code?
What once took hours of searching and a phone call to central office now takes seconds.
Tools like this are not about replacing expertise — they’re about augmenting it.
After a brainstorming session with our administrative team, I built a custom GPT — nicknamed The Methacton Guru — using ChatGPT-4o’s closed GPT capability through OpenAI’s Educator platform. The tool was trained on the following content:
- All district CBAs.
- Compensation and benefit plans.
- Local board policies.
- Administrative regulations.
- Pennsylvania school code.
- District safety procedures and protocols.
The build process required multiple testing rounds and formatting revisions to ensure the GPT was indexing information properly and producing accurate, aligned responses.
How It Works
From the user’s perspective, The Methacton Guru is intuitive. An administrator types a question such as, “How do I handle an employee misusing sick days?” and the GPT responds with synthesized guidance. This includes:
- The applicable clause in the relevant CBA.
- References to board policy and school code.
- Information on the grievance process (should the employee grieve your decision).
- Optional draft language for a formal letter.
The power lies in the tool’s ability to cross-reference multiple domains and present a clear, cohesive response. This functionality has been particularly valuable for questions involving process interpretation, policy enforcement, and employee management. What once took hours of searching and a phone call to central office now takes seconds.
Time Savings, Independence, and Trust
Our team immediately saw results. Leaders no longer need to call central office for basic interpretation of CBAs, compensation plans, or policies. Instead, they now have an assistant-level tool to guide initial decision making and reduce dependency.
To ensure reliability, I built the AI to prompt the user to contact me to verify AI outputs for the first three months. To date, we’ve had no incorrect responses.
Measurable benefits include:
- Dramatic reduction in policy-related inquiries.
- Faster decision-making cycles.
- Increased user confidence.
- Greater creative exploration of other AI use cases.
- We’re also fielding productive challenges from users: Can we add HR data? Safety forms? Financial protocols? It’s clear the tool’s value lies not only in what it does, but in what it inspires.
Security and Access
The tool is internal-facing and password-protected. It runs on OpenAI’s Educator AI license, which our regional consortium verified as highly secure — potentially more so than many internal legacy systems.
Administrators cannot edit or update the GPT directly; I maintain and update it centrally. When documents change, I simply swap in new training files, ensuring all outputs reflect the most recent version.
There are some technical limitations to be aware of:
- GPTs currently only accept 20 training files, so documents must be consolidated strategically.
- PDF and website content may need reformatting (e.g., BoardDocs does not translate well).
- All training files should be in clear, structured formats like Word.
Advice for Districts Looking to Replicate the Model
Districts considering replicating the model should:
- Define the goal clearly. Know what decisions or delays you're trying to improve.
- Choose the right documents. Prioritize foundational ones: CBAs, policies, comp plans, and code.
- Train and test thoroughly. Start with simple prompts; escalate to more complex ones to validate reliability.
- Manage expectations. This is not a human — it won’t reason like one. But it will dramatically improve information access.
- Secure access. Use a controlled, internal login system and audit usage regularly.
Scalability and the Road Ahead
The Methacton Guru can be replicated in any district, regardless of size. All it requires is access to your governing documents, a clear vision for use cases, and a willingness to test and revise.
We are now exploring separate GPTs for HR processes, financial protocols, and even district-wide summative student data analysis. Importantly, no identifiable student data is included in the training model. Where AI is concerned, ethics and security must lead.
Final Thoughts: The Iceberg Tip
AI in public education is still in its early stages. Tools like this are not about replacing expertise — they’re about augmenting it. Done right, a well-trained GPT becomes a smart assistant that empowers decision-makers with clarity, speed, and autonomy.
For us, the value has been clear: fewer interruptions, faster resolutions, and a more empowered administrative team. The Methacton Guru may be a local project, but its model is one all districts can build upon.