Welcome to the Ellen Ammann Dataset Manager. This application provides a simple, beautiful interface to help you build and manage the Knowledge Base and Evaluation Questionnaire for your upcoming RAG (Retrieval-Augmented Generation) system.
- Getting Started
- Session Management
- Managing the Knowledge Base (KB)
- Managing the Questionnaire (QA)
- Editing and Deleting
- Data Safety & Backups
- Troubleshooting & FAQ
- Open the Application:
- On Windows, double-click the
start.batfile in your main folder. This will automatically open a command prompt window (which runs the background server) and open your default web browser to the editor interface (http://localhost:3000). - On Linux/VM, use
start.shor run the Docker container as described in the README.
- On Windows, double-click the
- Switch Views: Use the toggle buttons at the top right to switch between the Knowledge Base (factual records) and the Questionnaire (evaluation Q&A).
The system supports multiple sessions, allowing you to manage different, independent datasets (e.g., your real dataset vs. a testing dataset) without them mixing up.
- Switching Sessions: Use the dropdown menu in the top right corner to jump between your active projects.
- Creating a New Session: Click the ➕ (New Session) button next to the dropdown. A secure dialog will ask for a name (use letters, numbers, hyphens, or underscores). This creates an isolated workspace for your data.
- Importing a Session: Click the 📂 (Import Session) button to upload an existing
.jsonlfile from your device into a new session workspace.
The Knowledge Base is where you enter the factual data about Ellen Ammann. This data is saved in sessions/<your-session-folder>/kb.jsonl.
- Click + New Record to clear the form.
- Fill out the Required Fields:
- Record ID: A unique identifier for this fact (e.g.,
ea_fact_0105). - Record Type & Category: Select an existing option or type your own custom category. The system learns and auto-suggests new categories!
- Subject: Usually "Ellen Ammann".
- Text: The factual sentence you want the AI to retrieve.
- Source IDs: Where did this fact come from? (e.g.,
src_ndb_1953). Use commas if there are multiple.
- Record ID: A unique identifier for this fact (e.g.,
- Click Save to KB File.
The Questionnaire tests how accurately your RAG system can answer questions based only on the Knowledge Base. This data is saved in sessions/<your-session-folder>/qa.jsonl.
- Click + New Question.
- Provide a unique Question ID (
Q01), the Question string, and the Ground Truth Answer. - Link to Evidence: In the "Supporting Record IDs", type the ID of the KB record that proves this answer is correct (e.g.,
ea_fact_0002). This is critical for evaluating whether the AI found the right evidence. - Click Save to QA File.
To Edit:
- Click on any record in the list on the left side of the screen.
- The form will automatically populate with that record's details.
- Make your changes and click Save. Because the system recognizes the Record ID, it will overwrite the existing entry.
To Delete:
- Click on an existing record to load it.
- A red Delete button will appear next to the Save button.
- Click Delete. A confirmation box prevents accidental deletions.
The Dataset Manager has a built-in "Ring-Buffer" safety mechanism to protect against accidental edits.
- Every time you click Save or Delete, the system takes a snapshot of your database and saves it in the
sessions/<session-name>/backups/folders. - These folders keep your 10 most recent actions. This acts as an immediate "undo"—if you accidentally mess up a record, simply open the backups folder, find the snapshot from 5 minutes ago, and copy its contents back into the main
.jsonlfile.
Warning
Manual Long-Term Backups
Because the system only saves the last 10 actions, a heavy session of editing will quickly overwrite your morning's work. At the end of every day, click the Download (↓) button in the UI or manually copy your files to a separate "Permanent Backups" folder on your computer!
Q: The page isn't loading or says "Site cannot be reached".
- Make sure the server script (
start.batorstart.sh) is currently running in your terminal/command prompt. Do not close that black window while working.
Q: I created a session, but my old data is gone!
- Your data isn't gone; you are just looking at a fresh workspace. Use the session dropdown menu in the top right to switch back to your original session.
Q: How do I completely start over?
- To reset a session entirely, you can either create a new session via the ➕ button, or close the server and run
reset_database.batin your file explorer to wipe the original demo databases.