Top10k
A research tool for planning software tutorials.
Top10k helps me review YouTube research, choose topics for Pixel & Bracket, and keep track of what I plan to record.
The project
Turn video research into a plan for what to record.
I run Pixel & Bracket, a business teaching people how to use software through tutorials. Part of that work is researching what people want to learn. Top10k is an internal web app I built to review other YouTube tutorials, compare their performance, and keep track of topics I might cover myself.
It sits on top of a database that collects information about videos and observes their view counts over time. The interface lets me narrow that collection, review candidates, and preserve my choices. It exists because a highlighted spreadsheet was becoming difficult to use for both research and production planning.
- Narrow the research. Filter videos by criteria such as the software covered or video length.
- Choose topics. Review candidates, add notes, and save selected videos to a list as references.
- Track my work. Mark the related tutorial work Recording, Done, or Skip, then return to those decisions later.
My role
Designing for a workflow I use myself.
I defined the research questions, designed the interface and its behavior, and directed development with AI coding agents. These are tools that can write and change code from instructions; my responsibility was to specify what the product should do, inspect the results, and correct behavior that did not fit the work.
Being the user made the problem concrete. I needed to distinguish a video worth inspecting from a topic I had chosen to record. I also needed my notes and progress to survive the next research session. A larger collection alone would not solve either problem.
How I built it
A browser interface, then another way to reach the same work.
Top10k grew within my broader set of tutorial-research tools. I built its interface over the existing video database rather than giving it a separate collection to maintain. Filters, saved selections, notes, and production status made that data usable for day-to-day decisions.
I later extended the interaction to voice and text. I wanted to review one candidate at a time while walking, ask follow-up questions, and find my selections in the desktop interface afterward. The browser and the assistant therefore needed to read and update the same records.
I specified concrete checks for that behavior: moving to the next video should not mark it permanently skipped; changing status should preserve notes; finishing a session without choosing anything should not create an empty list.
Decision 01
Separate a search from a selection.
A saved view remembers filters, such as videos about a particular app. Its contents can change as new data arrives. A selected list remembers the actual videos I chose. Treating those as the same thing would let a data refresh change my planned work.
I kept them separate and made production status belong to the video record. A topic marked Done should still be Done when the same video appears in another view or list. That gives exploration, selection, and progress distinct meanings.
- Views
- What’s relevant?Filters that update with the data.
- Lists
- What did I choose?Selections that stay selected.
- Status
- Where does it stand?Progress shared across the work.
Decision 02
Distinguish past popularity from current interest.
A video with a large lifetime view count is not necessarily gaining attention now. I kept its lifetime daily average separate from recent view growth. The latter requires comparing at least two observations taken at different times.
That means a newly added video may not have a recent-growth figure yet. Showing that gap is more useful than presenting a historical average as current momentum. The numbers help me compare evidence; I still have to decide whether I can make a useful tutorial on the subject.
Decision 03
Make spoken actions leave reliable records.
A voice session introduces ambiguity. If the research refreshes, “this video” must still mean the item we were discussing. If I say “next,” I may only want to move on for now, rather than exclude that topic from future work.
I used stable video identities to keep the current item fixed, separated session-only skipping from permanent status changes, and limited undo to selections added during that session. Status updates also preserve existing notes. These boundaries let a conversation change the work without silently undoing earlier decisions.
Result & reflection
A usable bridge from research to selected topics.
Top10k has implemented saved views, selected lists, production status, and voice operations. In September 2026, I used it to export a set of Gmail tutorial topics. That is a concrete example of research becoming a set of choices I could carry forward.
The tool remains an internal product. I am still evaluating voice reliability and how well selections carry through into completed tutorials. The lesson so far is that collecting information is only one part of the job: the interface also needs to preserve what I chose, why I chose it, and what happened next.