I was scrolling through Steven Soderbergh's SEEN, READ log on Extension 765 when I saw it. Below Deck. The reality show about drama on luxury yachts. Not once, not twice, but logged consistently from 2018 through 2023, spinoffs included, 35 episodes in 2021 alone. Right there in the log of one of the most respected filmmakers in the world. I did a double take, started watching, and got completely hooked. So did he apparently, which makes me feel a lot better about my viewing habits.
Soderbergh is the director behind Traffic, Erin Brockovich, and the Ocean's films. He has been sharing everything he has SEEN and READ, every film, every book, every TV episode, on his public blog Extension 765 since 2009. That's sixteen years of one of cinema's most restless creative minds leaving a trail of cultural breadcrumbs for anyone curious enough to follow. Sixteen years. That's not a log. That's a portrait.
So naturally, I fed it to a machine learning model.
The idea was simple and a little silly: if you treat Soderbergh's SEEN, READ history as a sequence, one item after another, year after year, can you build something that predicts what he might consume next given a few recent entries? Not a giant production AI. Just a compact, playful end-to-end exercise in scraping, cleaning, and recommendation logic. Totally reasonable use of a weekend.
Here's what I pulled together for version one. I scraped all 17 yearly SEEN, READ posts (2009 through 2025), parsed them into a structured activity log, normalized the item names so the model could find patterns, and trained a lightweight transition recommender. Basically: given these recent items, what tends to come next in the sequence? Then I wrapped it in a simple local web app so you could actually query it. I may have over-engineered the part where I gave it a name.
The dataset ended up being surprisingly substantial for something so niche: 6,145 total logged items, 3,836 unique ones. At one point I had to ask myself: am I actually a Soderbergh fan? So I checked his full directing portfolio and realized I've only watched a fraction of it. Which means I now have a very long watchlist and zero excuses.
The results are fun. Sometimes the model surfaces genuinely Soderbergh-coded suggestions, the kind of thing you'd believe he'd put on. Sometimes it's charmingly wrong in ways that reveal something interesting about the data. Either way, it's more entertaining than a standard "you might also like" widget.
Want to see what it recommends? You can try the tool yourself right here: Soderbergh ML Tool (local demo, instructions included).
But honestly, the most fun part of this project was just spending time with the dataset, watching the shape of a person's creative appetite emerge from a list of titles over sixteen years.
Soderbergh consumes culture the way the rest of us breathe. Meanwhile I'm over here watching Below Deck and calling it research.
This is an unofficial fan project built out of genuine admiration. All source material comes from Steven Soderbergh's public SEEN, READ entries on Soderblog at extension765.com. No commercial use. No affiliation. Just a fan with a laptop and too much curiosity.