PowerSchool Langflow

PowerSchool Langflow example is a project designed to integrate PowerSchool data with Langflow for generating text embeddings and building a vector store RAG (Retrieval-Augmented Generation) pipeline. The project leverages various tools and frameworks such as Conda, PowerSchool Adapter, Ollama, Langflow, and Chroma to facilitate the processing and storage of student data embeddings.

Perquisites

• Conda
• PowerSchool Adapter
• Ollama
• Langflow
• Chroma
• Ubuntu Server 22.04 or higher
• GPU is recommended for faster processing (RTX 4090 24GB or A6000 48G is recommended)

Installation

You can follow the instructions below to install the required components on your local machine or a server. Each component may have different operating system requirements.

1. Follow instructions here to install Ollama
2. Follow instructions here to install Langflow
3. Follow instructions here to install Chroma

---

Getting Started

Conda Environment

Create a new Conda environment:

> conda create -n powerschool python=3.12
> conda activate powerschool

Install the required packages:

> pip install -r requirements.txt

Now a new virtual environment is ready for this project.

---

PowerSchool Plugin

Install the plugin.xml.zip into your PowerSchool server and get the client ID and secret from the PowerSchool Plugin management. Feel free to extract the plugin.xml file and modify it to suit your needs. All the database fields defined in the plugin are read-only.

⚠️ Caution: Keep your CLIENT ID and CLIENT SECRET secure. Do not share them with anyone.

---

Environment

You can download this repository or fork it to your own repository. Create a .env file in the root directory of the repository and add the following environment variables:

POWERSCHOOL_SERVER_ADDRESS=https://your_powerschool_domain
POWERSCHOOL_CLIENT_ID=029408f0-e8d1-xxxx-xxxxx
POWERSCHOOL_CLIENT_SECRET=f81d4123-3156-xxxx-xxxxx
EMBEDDING_API_URL=http://your_ollama_host/api/embeddings
CHROMA_DB_HOST=xxx.xxx.xxx.xxx
CHROMA_DB_PORT=8000

Replace the values with your own PowerSchool server address. You can get the client ID and secret from the PowerSchool Plugin management.

⚠️ Caution: If your installed plugin for this repository includes database fields with Write permissions, ensure that the target server for this operation is a PowerSchool test instance. This code bears no responsibility for any unintended consequences.

Run the Application

If you are evaluating the application with a test PowerSchool instance and there is no student data, you can run these two Python file to create dummy student data:

1. Modify the DUMMY_SCHOOL_NUMBER to your own school number in the enroll_dummy_student.py file. The script will create 10 dummy student records in the PowerSchool database.

> python enroll_dummy_student.py

2. Run run_this_after_dummy_enroll.py to validate the data, you should see 10 JSON records in the console.

> python run_this_after_dummy_enroll.py

---

Text Embeddings

There is a create_embeddings class in the embedding.py file that generates embeddings for the PowerSchool data. The create_embeddings class uses the Ollama API to generate embeddings for the PowerSchool data. It's using the nomic-embed-text model to generate embeddings. You can modify the model name to use a different model.

Example:

{'embedding': [-0.6020767092704773, -0.7206705808639526, -2.4430699348449707, -0.6081189513206482, -0.3681376278400421, 0.30644160509109497, 0.8774333596229553, -1.1189463138580322, 1.1225370168685913, -0.03398386389017105, -0.9951059222221375, 0.05440517142415047, 1.6257606744766235, 0.06402243673801422, -0.13153652846813202, -0.5484852194786072, 0.40664592385292053, -0.3808309733867645, ..., -0.6412163972854614]
Payload sent to API: {'model': 'nomic-embed-text', 'prompt': "{'name': 'Hess, Travis', 'grade': '4', 'student_number': 'xxxxx', 'gender': 'Female', 'enroll_status': 'enrolled', 'district_entry_date': '2025-xx-xx', 'entry_date': '2025-xx-xx', 'exit_date': '2025-06-30', 'parents': {'father': 'Timothy Aguirre', 'mother': 'Patty Holmes'}} {'2025': {'tuition': '250885 RMB', 'capital_fee': '13248 RMB'}, '2024': {'tuition': '218220 RMB', 'capital_fee': '20417 RMB'}} [{'type': 'Lego Kit', 'checkout_date': datetime.date(2024, 4, 8)}, {'type': 'Camera', 'checkout_date': datetime.date(2024, 3, 28)}, {'type': 'History Book', 'checkout_date': datetime.date(2024, 9, 8)}, {'type': 'Chinese Book', 'checkout_date': datetime.date(2024, 10, 10)}] {'HepB': {'1st': datetime.date(2007, 12, 16), '2nd': datetime.date(2010, 5, 22), '3rd': datetime.date(2019, 6, 19)}, 'DTaP/Tdap': {'DTaP_series_completed': 'pre-age 7', 'Tdap_booster': datetime.date(2022, 7, 17)}}"}

Run the Application

When you get all the required environment variables and the dummy student data is ready, you can run the application:

> python vectorize_student_example.py

It will read the student data from the PowerSchool database and generate embeddings for each student. The embeddings are stored in the Chroma database.

After the embeddings are generated, you can start build your vector store RAG in Langflow with these components:

Click on the Playground button to start the pipeline. You can interact with the pipeline using the Chat Input component.

Demo

https://assets.tonylabs.com/videos/powerschool/powerschool-langflow-rag.mp4

---

Ollama

Install Ollama and choose a LLM model here. The nomic-embed-text model is required to generate embeddings for the PowerSchool data.

Recommended LLM:

• llama3.2 1B and 3B
• mistral
• qwen2.5 3B and 7B

---

Langflow

Drag and drop a Chroma DB component with the following settings:

• Collection Name: students
• Persistence Directory: /data/chroma
• Search Query: connect to the Chat Input component
• Embedding: connect to an Ollama Embeddings component
• Search Results: connect to aParse Data component

---

Vector Database

Chroma is recommended for storing and searching embeddings because it's full local and doesn't require an internet connection. It's safe and secure for storing sensitive data.

---

Chatbot

You can install your favourite Web UI chatbot to interact with the Langflow pipeline. The chatbot can be hosted on a server or locally on your machine. If you don't familiar with Web UI chatbot, you can use the Chat widget from Langflow. It's a simple and easy way to interact with the pipeline.