An official website of the State of Maryland.

Maryland.gov Translate

Official websites use .gov
A .gov website belongs to an official government organization in the United States.

Secure .gov websites use HTTPS
A lock ( ) or https:// means you’ve safely connected to the .gov website. Share sensitive information only on official, secure websites.

Digital Sandbox for Training

YouTube-to-Docs summary

Problem & Costs

State agencies, researchers, and compliance officers rely heavily on reliable text-based records of public proceedings. Public meetings, training videos, and legislative hearings are regularly hosted on YouTube. However, video content is subject to strict legal mandates for digital accessibility, requiring state entities to provide accurate captions, transcripts, and multi-language translations for limited English proficient (LEP) individuals. YouTube’s native, auto-generated captions are consistently inaccurate, regularly misinterpreting regulatory language, technical state acronyms, and speaker identities, while failing to provide reliable localized translations. Public servants are currently forced to dedicate extensive administrative hours to manually creating transcripts or reviewing faulty transcripts. Only after establishing an accurate textual baseline can staff begin drafting notes, indexing speakers, and extracting key insights. This manual workflow creates three specific problems for public-sector agencies:

  • High Labor Costs: A typical two-hour agency briefing takes hours of human labor just to transcribe and outline. Multiply this across dozens of videos a week, and agencies spend thousands of dollars in employee hours on basic administrative overhead instead of actual public service.
  • Poor Accessibility: The standard, auto-generated YouTube captions are frequently inaccurate. They often miss technical terms, agency-specific acronyms, or proper names, making the content inaccessible to individuals who rely on clean text summaries or assistive reading tools.
  • Information Bottlenecks: Important details stay locked inside video and audio files. Because there's no fast way to search through a hundred hours of video, policy analysts often miss critical context from past discussions.

Under current Maryland policy, all technology procurements, software adoptions, and platform approvals must clear the Department of Information Technology (DoIT) to ensure compliance with state security, data privacy, and budgetary guidelines. IT solutions for this bottleneck could include navigating protracted procurement cycles to acquire expensive third-party transcription software, or forcing each agency to make-do with a mix of AI assistants—resulting in cross-agency inconsistencies and possible data breaches.

Solution

DoIT is currently executing a broader cultural shift across Maryland state government, moving from a "buyer" mindset toward a "builder" approach for internal technologies. This transition has been underway to reduce reliance on costly external vendor contracts and to ensure the state retains direct control over its data security and software architecture. YouTube-to-Docs is a direct product of this ongoing shift. Rather than purchasing an expensive, third-party enterprise software package or allowing fragmented, non-compliant AI adoption, DoIT leverages its internal engineering capabilities to address the need for high-quality video transcripts.

Developed using agentic engineering principles, YouTube-to-Docs balances traditional software engineering disciplines (e.g., automated testing, strict type checking, linting, accessibility standards, manual verification scripts) with the accelerated speed and flexibility of AI innovation. In applying professional expertise to guide and audit AI-generated components, developers can minimize the manual effort of writing boilerplate code while maintaining strict software quality, security, and stability.

How it Works

YouTube-to-Docs is an open-source, cloud-, AI-, and vendor-agnostic utility designed to meet developers where they're at. It runs on a local machine or deploys across any infrastructure (including AWS, GCP, and Azure) with native storage integration for Microsoft SharePoint, Google Workspace, or local folders. The repository provides three distinct interfaces to fit existing agency workflows: a web UI featuring an intuitive user interface, an API for programmatic transcript extraction, and an integrated Model Context Protocol (MCP) server that exposes the tool natively to AI environments so users can extract transcripts without using a Command Line Interface (CLI).

The tool accepts video URLs, unique IDs, playlists, channel feeds, and comma-separated batches. It pulls raw audio or existing transcripts directly from YouTube, utilizing a bidirectional speech-to-text and text-to-speech pipeline to handle multi-language translations. Instead of spending hours manually reviewing video content, state agencies can turn any YouTube video into secure, searchable documents in under two minutes. Once processed, the data is saved directly to the designated workspace, where an interactive follow-up interface allows users to query the material and extract specific context on demand.

Scope of Use

The YouTube-to-Docs tool is built specifically for public-use, open-source video and audio content. Because it is designed to process public YouTube links, its scope is strictly limited to information that is already available to the general public. Approved use cases include processing public legislative hearings, town halls, governor press conferences, public agency training videos, and educational webinars. Conversely, state employees cannot use this tool for internal, confidential, or sensitive state data. It is explicitly prohibited for closed-door meetings, private personnel discussions, or any video containing Protected Health Information (PHI) or Personally Identifiable Information (PII) that is not already legally cleared for public viewing.

This utility is strictly a developer tool, and the web UI interface is meant for agency developers and system administrators rather than production-level public deployment. Because major cloud service provider IP ranges are systematically blocked from crawling YouTube, the tool is not designed to be deployed as a public cloud-hosted service. Running the application locally or within internal agency infrastructure avoids these IP restrictions, secures the handling of OAuth client IDs and client secrets, and strictly limits the network exposure of necessary application API keys.

Policy Compliance

YouTube-to-Docs is fully compliant with Maryland’s guidelines for responsible AI use as managed under Governor Wes Moore’s Executive Order on Artificial Intelligence. Because the utility only processes public YouTube links containing Level 1 Public Data, it does not put sensitive state networks or information at risk.

  • Data Privacy: The tool uses enterprise-grade API connections. This means that any data sent to the AI model for cleaning and formatting is processed within isolated enterprise boundaries and is never used to train public AI models, keeping Maryland's processing workflows secure.
  • AIB Alignment: 
  • Accessibility Standards: By turning messy, auto-generated video captions into clean, highly accurate text transcripts, the tool actively helps state agencies meet digital accessibility requirements for individuals with hearing impairments or learning disabilities. The web UI explicitly utilizes the Maryland Web Design System (MDWDS) library to ensure that the interface itself complies with state usability and accessibility standards out of the box.
  • Transparency and Verification: The tool acts as an administrative assistant, not a final decision-maker. In alignment with Maryland's Responsible AI Policy Implementation Guidance, state employees are strictly required to maintain human-in-the-loop oversight, verifying all generated content against the original video to ensure accuracy before using them in an official capacity.

Demo

To see YouTube-to-Docs in action, watch the video demonstration:

Watch the youtube-to-docs Video Demonstration

Implications & Next Steps

Moving forward, the next steps for this tool involve defining its long-term scaling strategy and integration into Maryland's broader AI framework.

  • Approved Tooling Only: Deployments must strictly adhere to DoIT-approved infrastructure, LLM endpoints, and vendor environments to ensure compliance with state security standards.
  • Cost Management: While the tool drastically reduces individual agency licensing fees, administrators must actively monitor API token usage and local infrastructure compute costs to prevent operational budget drift.