## [Screenshot]: SemanticCite Web Application Interface
### Overview
This image is a screenshot of a web application interface for a tool named "SemanticCite." The application is designed for AI-powered semantic citation verification, allowing users to analyze whether citations in their text are accurately supported by provided reference documents. The interface is divided into two primary sections: a left-hand sidebar for configuration settings and a main content area for the core citation verification workflow.
### Components/Axes
The interface is structured into two main vertical panels.
**Left Sidebar (Configuration Panel):**
* **Header:** "Model Configurations"
* **Section 1: LLM Configuration**
* Label: "LLM Provider" | Dropdown Value: "local"
* **Section 2: Dual Model Configuration**
* Label: "Preprocessing Model" | Input Value: "ollama/SemanticCite-Refiner-Qwen3-1B"
* Label: "Classification Model" | Input Value: "ollama/SemanticCite-Checker-Qwen3-4B"
* Label: "Model endpoint URL" | Input Value: "http://localhost:11434"
* **Section 3: Embedding Configuration**
* Label: "Embedding Provider" | Dropdown Value: "local"
* Label: "Embedding Model Name" | Input Value: "Qwen/Qwen3-Embedding-0.6B"
* **Section 4: Disclaimers**
* A highlighted box contains the text: "Privacy: All processing occurs locally"
* Below the box: "AI-assisted analysis - verify results independently"
**Main Content Area (Workflow Panel):**
* **Header/Logo:** A shield icon with "AI" and the text "SemanticCite".
* **Title:** "AI-powered semantic citation verification"
* **Subtitle:** "Analyse whether your citations are accurately supported by reference documents."
* **Section 1: Citation Text**
* Label: "Enter citation text"
* A large, empty text input area.
* **Section 2: Reference Document**
* Two tabs: "Upload File" (selected) and "Download from URL".
* Label: "Upload reference document"
* A drag-and-drop file upload zone with the text: "Drag and drop file here" and "Limit 200MB per file • TXT, PDF, MD". A "Browse files" button is on the right.
* **Section 3: Reference Metadata (Optional)**
* Label: "Enter reference metadata"
* A text input area with placeholder text: "Optional: Enter title, abstract, and other reference metadata... Supports plain text and markdown formatting."
* **Action Button:** A large blue button at the bottom labeled "Check Citation" with a magnifying glass icon.
### Detailed Analysis
The interface is a form-based application. The user is expected to:
1. Configure the underlying AI models (LLM and Embedding) in the left sidebar, which are pre-filled with specific model names and a local endpoint.
2. Input the citation text they want to verify in the "Citation Text" field.
3. Provide a reference document by either uploading a file (TXT, PDF, MD, up to 200MB) or downloading from a URL.
4. Optionally, provide additional metadata about the reference document.
5. Click the "Check Citation" button to initiate the analysis.
All configuration fields in the sidebar have a help icon (a question mark in a circle) next to their labels. The main content area's input fields also have these help icons.
### Key Observations
* **Local Processing Emphasis:** The configuration defaults to "local" for both LLM and Embedding providers, and the endpoint URL points to `localhost:11434`. This is reinforced by the prominent privacy disclaimer stating "All processing occurs locally."
* **Specific Model Stack:** The application is configured to use a specific suite of models from the "Qwen3" family, hosted via "ollama": a Refiner model (1B parameters), a Checker model (4B parameters), and an Embedding model (0.6B parameters).
* **Dual-Model Architecture:** The "Dual Model Configuration" suggests a pipeline where one model preprocesses or refines the input, and another performs the core classification or checking task.
* **User Guidance:** The interface includes placeholder text and disclaimers to guide the user, such as the note to "verify results independently" and the metadata field's explanation of supported formatting.
### Interpretation
This interface represents a specialized tool for academic or research writing integrity. Its purpose is to automate the laborious task of cross-referencing citations against source material to detect inaccuracies or misrepresentations.
The design prioritizes transparency and control. By exposing the specific model names and the local endpoint, it caters to technical users who may want to understand or modify the underlying AI stack. The strong emphasis on local processing addresses potential privacy concerns, assuring users that sensitive documents and citations are not sent to external servers.
The workflow is linear and logical: configure models, provide inputs (citation + reference), and execute the check. The optional metadata field suggests the tool might use this additional context to improve verification accuracy. The "Check Citation" button is the clear call-to-action, positioned as the final step in the process.
Overall, SemanticCite appears to be a focused application leveraging local, specialized AI models to provide a privacy-conscious solution for verifying the fidelity of scholarly citations.
