Back in February, I wrote about the playlist checker tool I had built with Claude that helped us monitor YouTube links in our Notion database. In that, I referenced the elusive "Official Archive" that we've mentioned over the years.
Freaking years. That's what it's been. I've rather consistently pushed this project over the horizon, constantly thinking if I just learn a bit more about Python, a bit more about Django, and submit myself to a bit more web development Tutorial Hell™, I'll eventually be ready to embark on the Official LCC | SaberComp Archive 9000™. Oh and in the meantime I must surely prototype everything possible in Notion.
Well after finally finishing my project of rebuilding my personal site from scratch on Astro, I thought about my next Claude-Code-assisted project. Regarding the Tutorial Hell™, I had previously been working my way through Writing [My] First App with Django, the Python web framework I've long had selected for the SaberComp Archive project. But with the general goings-on of life, said Tutorial Hell fell by the wayside. And one of my Claude testbed projects was having it look at that codebase and remind me where the hell I had left off.
So instead of once again embarking on figuring out how to code things one line at a time, I decided in the interest of actually shipping things, I would trust Claude enough to do the legwork for me. This would instead be a directing job for me, something I could even let churn while I tended to actual work work.
So a couple weekends ago, I created a rather large prompt, along with a handful of links to various web sources for Claude to follow, and my ideas of how I envisioned this whole archive to work—essentially an IMDb for SaberComp.
Within 20-30 minutes, it had a very basic thing up and running.
Freaking thing scraped two* web sources of entries, had some preliminary styling for the front end (and I mean preliminary; basically just confirming stuff can be published on the front end at all), and had some basic back-end admin control panel stuff. That's actually a cool attribute of Django—you get admin dashboards for damn-near free.
Feeding the Database
This was a cool part that I had long pondered and Claude just... did it. The entirety of LCC | SaberComp is scattered across the internet in various forms. The original Unofficial LCC Archive on PhiSpace. The Notion database I've been filling out, and the YouTube Playlists Christopher Spenceley has been building in recent years. All of these became insanely useful.
And Claude made them useful by building custom importers for each. This was something I knew would need to be done, and I knew I'd be able to learn it eventually through the various Python books and training sources I had... It's basic web scraping. But we're in the age of AI coding, so let's just freaking do this shit and be done with it.
*I mentioned two sources above, but then listed three. I would be remiss if I didn't mention how part-way through all this, Claude noticed that we seemed to be missing a lot of entries, at which point I fed it the YouTube playlists. Not only did it go "ooo, more data!", but it was able to notice overlap, notice naming conventions ("LCC III Entry 01 - The River"), and prepare the importer to not duplicate existing entries, and note what was a title vs entry number, as early LCC did.
Tables. Tables Everywhere
Something that happened early on that ended up being a huge blessing later is that I took great pains in explaining minute ways I envisioned the archive to function from a user standpoint. This in turn made Claude suggest certain database structures... which in turn made things more complex, but more flexible down the line.
We have about 22 tables within the database, things like Video, Person, Credit, Film, Entry, Channel, Role, Accolade, Submission, etc, et-cet-freakin-a. It sounds insane. But time and time again, as I've asked Claude about building out some new complexity, it's come back with "yeah that's totally possible, because we built it this way."
Reusing Tools
Having had other Claude Code sessions build out Python tools to do other things, I pointed Claude at that directory for reference. I don't think I've shared it here on the blog but we have a particularly slick one that looks at the raw credits text field that we ask for in the submission form, sends that out to the Claude API, returns it as JSON data, and plugs it back into the Notion database.
Welp, we also now have that tool built right into the Django archive. Raw credits, either imported from Notion or plugged in by us (or the future submission form), can be parsed by the Claude API and immediately create people in the database, match to existing people, and match to the closest existing film role. Freaking nifty, as the kids say.
There are a metric buttload more cool things happening under the hood. We'll talk about that eventually.
Keeping Humans in the Loop
One of the most important things about this whole archive effort is accuracy. Constantly along the development effort with Claude, the human factor has been stressed. Whenever something can possibly be figured out with the Claude API, or a Python tool, the final approval step is human-driven. While we might scan your credits with Claude, or scrape it from the internet, nothing ends up on the front end of the archive site without human approval.
A lot of the directing process on my part has been deciding how we classify something that comes from one source, how we classify it when it's AI-created, and how we indicate something is good to go. Then how do those various statuses get displayed, or not displayed, or not displayed and reported... on the front end.
All this to say... despite Claude Code doing a tremendous amount of coding work on this, I still have a very deep hand in guiding its development. There are extensive decision-making processes along the way... even simple things like "how does the 'person duplicate' scanning tool scale over time?"
One of the most intense things that has happened across this development process, I realize is something that I would've botched endlessly, and to great frustration, if I had been trying to do this without the help of Claude. Every time I realize we need to change some tables around, or completely rethink how we're tracking something, the database has to be either rebuilt or edited in a very careful way.
There was a night where we had to completely rethink a few things, and it turned into a 5-stage plan from Claude where each piece became about 30 minutes of back-and-forth decision-making. And each stage had to be done in a precise order so that all the data would be preserved through the migrations.
On top of this, Claude has something like 500+ tests it's running every time it makes a major change at this point. This is an area where I've never been proficient in my development work. And it's wild to see this AI system just preemptively do it. I don't know if it's a matter of Claude picking up that I know just enough... or it realizing I'm an idiot.
What's Next
At the moment I'm in a bit of a back-end stress-test. I'll go about doing some cataloguing; running the credits parsing tool, confirming things, adding roles, checking the front end, and going back to Claude when I find workflow issues that could be smoothed out.
Medium-term, there's the front-end styling, which will be fun and involve a lot more input from the SaberComp Council. With SaberComp 2026's deadline approaching, I plan on testing our submission process by personally plugging in everyone's entries as a sort of mock competition, seeing how the submission form + archive works in real time.
Long-term, I'm hoping we can launch a public archive in time for the 2027 season, so that we can use the fancy submission form—whereby your entry will land on the back end in an envelope for our review, and upon approval, be entered in the database automatically. This tooling is already working, but as I just alluded, has not been fully tested.
Your Help
Beyond that, when we do launch, I intend to have some simple tools for the community to help with the archive. Namely—a "submit a revision" link on any given page where you can tell us what we have wrong, or information you would like to see added.
My vision of the LCC | SaberComp Archive being a permanent home for 20+ years of lightsaber goodness is far closer than I ever envisioned. I'm pretty psyched to share more.
Other than all of that, I won't say no to monetary help to allow me to justify plugging away at this stuff like a crazed raccoon.
Current Stats
- 24 Competitions
- 519 Entries
- 641 people so far