DepthChartIQ program pages
Public profile pages for D1 men's and women's basketball programs on DepthChartIQ.
LSU, Auburn, Memphis, Princeton and Pacific are all Tigers.
So the model that checks each logo has to read the school's name on it. A tiger alone won't settle it.
I built the pipeline that looks for each program's logo, coach photo and arena photo, and checks what it finds.
A search model finds the right Wikipedia page for a school, and that's all it's trusted with. In a test on five schools it made up the image addresses nearly every time, so plain API calls fetch the real file and its license instead. The pipeline copies an image onto DepthChartIQ's own storage only when its license allows reuse, and it looks for coach photos only in sources that allow reuse.
Then a small, fast vision model reads each logo, words and all. If it isn't confident, a bigger model takes a second look.
The week it went public
On August 10, 2026, I told the CEO the public pages would go up. They went live between August 15 and 18.
The pipeline already existed. It dated from May, when its first full pass took about 12 hours over 22 runs, cost $7.27 and shipped zero logos the checker wasn't sure of. So on paper, the week was just filling in the gaps.
The gaps took 73 pull requests, which I merged between August 12 and 15: the public pages themselves, the women's programs, arena photos and coach photos.
The CEO rebuilt the program profile as one page for signed-in and public visitors, and added the directory search. His notes on the search say "st johns" found nothing, because the apostrophe in "St. John's" split the name in two. He fixed that and added an alias table of 100 schools so "UNC", "Ole Miss" and "UMass" find their programs.
Which leaves the logos that are just a tiger or a bulldog, with no name anywhere on them. Those need two of three other checks to agree, such as the school's colors.
Otherwise, a person looks at the tiger.
Stack
- Data
- Cloudflare R2
- Infrastructure
- Inngest