YouTube to docs: how Maryland is taking videos of public hearings and turning them into usable written content

The problem

State legislatures and government agencies hold public meetings, bill hearings, and board sessions that can run anywhere from three to seven hours. These are streamed live on YouTube to make them publicly accessible, and the recordings remain available afterwards. But the length of these sessions makes it difficult for anyone — whether a member of the public, a government employee, or a legislator — to go back and find out what was said, who said it, and what decisions were discussed. Unless someone sat through the entire session, there was no practical way to get a clear picture of what happened.

The need for a solution surfaced through the state's IT software review board, where agencies submit requests to purchase technology. One agency — the Accountability Implementation Board, which analyses school effectiveness across the state — wanted a tool that could help summarise its meetings and produce meeting minutes from YouTube recordings. The state's AI and Machine Learning Product Director recognised that the core task — pulling information out of long YouTube videos and making it useful — could be built internally rather than purchased, using freely available tools and AI services.

The solution

YouTube to Docs is a free tool that takes YouTube videos of public meetings and hearings and turns them into usable written content. Rather than procuring a commercial product, the tool was built in-house and made publicly available on GitHub, a platform where software code is shared openly so that others can use, adapt, and improve it.

How it works
Every YouTube video has an automatically generated transcript — a written record of everything said in the video. YouTube to Docs pulls this transcript and then uses a Large Language Model (LLM) — a type of Artificial Intelligence (AI) that can read and analyse large amounts of text — to process it in several ways.

Users can get a raw transcript with timestamps, which can also be used to create captions for videos. They can generate a structured summary of the session, condensing hours of discussion into a short readable document. They can extract who spoke and what they said, or pull out the questions that were asked and the answers that were given. They can also translate the content into other languages.

For users who prefer to work with their own tools, YouTube to Docs can also simply provide the raw transcript to paste into whichever AI assistant they already use — such as Copilot, Gemini, or Claude — and ask their own questions about what was discussed.

How people use it
There are two ways to use YouTube to Docs. The first is through a command line — a text-based interface on a computer where users type instructions rather than clicking buttons. A user types a single command along with the link to a YouTube video, and the tool does the rest. The second is through a simple web page that YouTube to Docs can generate on the user's own computer, which provides a visual interface for selecting options such as which AI model to use and whether to translate the output.

What it works with
The tool was designed to work with whichever technology an organisation already has access to. It is compatible with Google Workspace, Microsoft SharePoint, Amazon Web Services, Google Cloud, and Microsoft Azure. It also supports free AI models, such as Google's Gemma 4, meaning the entire process can be run at no cost.

How it was built
The tool was built using AI coding assistants to help write the initial code — particularly for the more complex parts that connect to different cloud services — while the developer directed the process and checked the output as it progressed.

The first version was deliberately simpler — a Google Colab notebook, an online platform where users can run code in a web browser without installing anything on their computers. This made it accessible to people with no technical background and allowed them to try generating transcripts without needing to learn any new tools. The current version is more powerful but requires users to install the software on their own devices and use a command line.

The impact

YouTube to Docs has been used to process bill hearings and board meetings across the state. During the most recent legislative session, it was used to analyse hearings where the Department of IT had bills under consideration, producing summaries of the questions asked, the answers given, and the key points of discussion.

Download statistics show approximately 150 downloads per day, and users outside the state have also reported independently adopting the tool.

Specific time savings have not been formally measured, but the practical value is straightforward: a hearing that would take hours to re-watch can be turned into a summary and searchable transcript that takes minutes to review.

Lessons learned

• Government can build, not just buy. The tool was built in-house at no procurement cost, using freely available code libraries and AI models. This challenged the common assumption that government always needs to purchase technology from a vendor. AI tools have lowered the barrier to building internal solutions, and this project shows what a single technically skilled public servant can produce in a short period of time.
• Start simple to build familiarity. The first version was a simple online notebook that required no installation and no prior technical knowledge. This helped people who had never written code become comfortable with the concept before the tool moved to a more capable but more technical format. For teams introducing new tools, offering a simplified starting point can help build confidence and encourage adoption.
• Technical barriers to adoption are real. The current version requires users to work with a command line — a way of interacting with a computer by typing text commands rather than using a visual interface. For people who have never used one, this can feel unfamiliar and intimidating. This is a genuine challenge for any tool aimed at government staff who may not have a technical background, and it is something the project continues to work on.
• Design for flexibility. By building the tool to work across multiple cloud providers and AI models, the developer avoided tying it to any single platform. This matters in government environments where different agencies may use different systems, and where procurement rules can limit which services are available.