Skill UI

A comprehensive workflow for executing advanced, semantic web searches using Exa. This tool moves beyond simple keyword matching, enabling deep context retrieval by understanding the intent behind the query. Key features include RAG context retrieval, content extraction (summaries, highlights, full text), and robust filtering by date range, specific domains, and content categories. Ideal for building knowledge base search features and complex research pipelines.

MarkItDown is a lightweight utility designed to convert heterogeneous documents (PDF, Office files, HTML, etc.) into structured, plain Markdown. It is specifically optimized for text analysis, search indexing, and preparing data for Large Language Models (LLM) and Retrieval-Augmented Generation (RAG) systems, prioritizing content structure over visual fidelity. Supports local, stream, and advanced multimodal extraction scenarios.

This comprehensive guide covers essential operations for handling PDF documents, including text, table, and metadata extraction using libraries like pdfplumber. It details document creation with reportlab, merging/splitting with pypdf/qpdf, and advanced OCR for scanned PDFs. Ideal for automating complex document workflows and structured data ingestion.

This powerful router analyzes complex document types (scanned, medical, table-heavy, etc.) and determines the optimal conversion pipeline (tools, flags, cleanup) to extract data into various formats like Markdown, JSON, or DOCX. It goes beyond simple text extraction by prioritizing the preservation of document structure, relational data, and high fidelity, ensuring the generated output is accurate for downstream machine processing.

LiteParse is a fast, open-source, and local document parser designed for layout-aware text extraction from PDFs, images, and office files. It outputs structured JSON with bounding boxes and metadata, making it ideal for building advanced RAG systems, citation grounding, and complex data pipelines without needing cloud APIs. Features include OCR, batch processing, and page screenshot capture.

This skill allows users to extract structured data (text, tables, attributes) from rendered web pages. It enhances traditional web scraping by ensuring all extracted strings pass mandatory AIDefence PII checks and prompt-injection gates before being stored or returned to the model. It supports building reusable extraction templates and running one-shot queries.

A comprehensive skill to extract and transcribe subtitles from any YouTube video URL. It utilizes a two-stage approach: first, attempting fast extraction via the yt-dlp CLI tool; failing that, it falls back to advanced browser automation using Chrome DevTools MCP to scrape the transcript panel content directly from the webpage DOM. The output is always a clean, timestamped plain text file.

This skill facilitates the extraction of population-scale (cohort-level) clinical text from Electronic Health Records (EHR) or FHIR data warehouses using the FHIR Bulk Data Access ($export) operation. It manages the asynchronous job kickoff, polling, and downloading of NDJSON files. The resulting data stream is designed to be fed in batches into OpenMed for mandatory de-identification and Named Entity Recognition (NER) analysis.

This utility separates clinical entity extraction from clinical coding. It processes de-identified text, utilizing OpenMed to identify key clinical spans. These spans are then programmatically mapped into structured FHIR R4 resources (like Condition, Observation, and MedicationStatement) and assembled into a Bundle. This process ensures deterministic, standardized data output suitable for interoperability and medical record structuring.