KeywordExtract is a specialized research tool designed to ingest academic PDF papers and extract structured data regarding object affordances, robotics, LLMs/VLMs, and safety using Google's Gemini 3 Flash model.
It automates the literature review process by running versioned prompt templates against uploaded papers to answer specific research questions (RQs).
- PDF Ingestion: Drag-and-drop support for uploading multiple academic papers (PDF format).
- Gemini 3 Flash Integration: Uses the latest multimodal models to parse complex academic text and extract semantic concepts.
- Structured Extraction: Runs defined "Prompt Templates" corresponding to specific Research Questions (e.g., "RQ2: Planning Role", "RQ3: Datasets", "RQ4: Safety Measures").
- Batch Processing: Analyze multiple papers against a selected prompt in parallel.
- Export Data: Download extraction results as JSONL (for programmatic use) or CSV (for spreadsheet analysis).
- Prompt Library: A curated list of extraction prompts tuned for affordance learning and embodied AI research.
- Upload: Go to the Paper Vault and upload your PDF collection.
- Select Prompt: Navigate to the Prompt Library to choose a specific extraction goal (e.g., "Extract Dataset Limitations").
- Run Analysis: The system processes the text of each paper through the Gemini API.
- Review & Export: View the extracted keywords, quotes, and confidence scores in the Results view, then export to CSV/JSONL.
To run this application on your own machine (to avoid cloud timeouts), see README_LOCAL.md.
The tool is currently configured to extract information related to:
- RQ1: Definitions of Affordance.
- RQ2: Role of affordances in Planning, Decision Making, and Imitation Learning.
- RQ3: Architectures, Datasets (construction & limitations), and Prompting Strategies for VLMs.
- RQ4: Hallucinations, Error Detection, and Safety Measures in affordance-based systems.
