What is this even?
Hi, I'm Tim and I've been dreaming about building VaporLens for the past decade. The idea was born after I listened to one of the rants by the great TotalBiscuit (RIP - I still miss his content) about how game scores are meaningless without knowing the reviewer's preferences. After all, if you don't know what someone likes, how can you tell if their score is relevant to you?
Since then, I've been pondering whether there's a better way to present video game reviews in a concise and simple format. After more than a decade of trying to figure this out - without spending a fortune on NLP tool R&D - I finally cracked it in 2025 with the help of modern LLMs.
This is my passion project and the culmination of decades spent obsessing over this one idea. I personally find it incredibly useful and plan to keep improving it for myself as long as I can.
If you find it helpful, share it with friends. Seeing those numbers go up makes my data-driven soul happy :)
And if you'd like to support me directly, there are a few ways to do that below.
VaporLens uses modern LLMs to efficiently analyze Steam reviews. It starts by sampling up to 2000 high-priority reviews, skipping the need to process hundreds of thousands. Reviews are weighted by recency (posted after the latest patch), depth (length), player experience (playtime), and community upvotes to ensure quality insights.
The system then filters out unhelpful content (memes, review bombs, short comments), extracts key statements about positives, negatives, gameplay, performance, and recommendations, and summarizes them into concise overviews. The result is a clear, noise-free snapshot of player sentiment - no endless scrolling required.
Yes, it's free. And no, I will not add ads ever. I hate ads.
I might add extra paid features or affiliate links to stores like Fanatical at some point, but I'll try to keep them non-invasive.
The core experience will always be free.
I can currently process ~6,000 reviews per day. It costs ~$0.15 per 1,000 reviews. And yes, I'm covering that cost myself right now.
Steam's API only allows fetching reviews from the last 365 days - and even then, it doesn't always return everything (even with filters set to "all"). We work with what we get! That is also part of the reason I've introduced sampling.
It includes all reviews Steam provides, regardless of language. That's the beauty of LLMs - they don't particularly care about the language used in the original review.
Importance is a relative score assigned to each point, mostly based on frequency but also factors like relevance, helpfulness, and how emotional/clickbaity it is. Generally, higher importance = more mentions.
This includes any statement about recommending the game, from explicit “I (don't) recommend this” to suggestions like “wait for a sale” or “wait for patches.”
Most are removed during initial filtering. If any slip through, it means they contained something meaningful. Overly emotional reviews or baseless attacks on developers are deranked (lower importance), so you'll only see them in the Stats view.
VaporLens uses embeddings from all feature categories to create a composite game embedding. These embeddings are then compared to determine the "closest" matches. Generally, this produces solid results, but you might occasionally see less relevant suggestions - this typically occurs due to a limited number of games in the current database. VaporLens simply selects the 10 most similar games available. When there aren't enough truly comparable titles in the system, you'll see less ideal recommendations. This temporary issue will naturally resolve as the game library expands.
Maybe someday. Right now, it's a bit of a mess 😅
Awesome! Send them my way - you can find my contact details here.