🚀 I built Brickbeam — an AI-powered assistant that helps LEGO fans turn their messy piles of bricks into real builds.
If you have kids, you probably know this situation: hundreds (or thousands) of LEGO pieces mixed together in a box, and nobody knows what can actually be built with them. 😅
I created Brickbeam to solve this problem.
📸 Take a photo of your LEGO bricks
🤖 AI analyzes your pieces
🧱 Discover sets and builds you can create with what you already own
The idea came from a simple frustration: LEGO collections keep growing, but many pieces end up forgotten at the bottom of the box.
Instead of buying new sets, what if you could rediscover the possibilities hidden in your existing bricks?
Brickbeam is now live on the App Store:
https://apps.apple.com/tr/app/brickbeam/id6794062762?l=en
I’d love to hear your thoughts:
Would you use an app like this for your LEGO collection?
What features would make it more useful?
Any LEGO fans here who want to try it?
This is my first step in building Brickbeam, and I’m excited to share the journey with the Indie Hackers community. 🚀
This is a really cool use case for AI.
The “forgotten pile of LEGO” problem is super real, especially for families — there’s a lot of unused value just sitting there.
I like that you’re not trying to sell more sets, but helping people get more out of what they already have.
One thing that could make it even stronger: showing progressive builds (start with what you have → suggest what small extra pieces unlock bigger builds).
Curious how accurate the detection is with mixed/partially hidden pieces — that seems like the hardest part to get right.
Great idea, feels very practical.
building something is not easy that's look so funny i'm gonna try
thank you.
great
How are you positioning against Rebrickable, which already owns the "what can I build from my parts" behavior with AFOLs? Your wedge is probably the camera: if scanning a messy pile is 10x less friction than cataloging parts by hand, lead every demo with that one moment. The LEGO communities are dense and vocal, so win one subreddit or forum and word of mouth does the rest.
That's exactly how I think about it.
Rebrickable is an amazing product, especially for dedicated LEGO fans who already maintain detailed inventories. My goal isn't to replace that workflow—it's to make the first step dramatically easier.
Most casual LEGO owners don't have the time or patience to catalog thousands of pieces by hand. If they can simply dump their bricks on the floor, take a photo, and immediately discover what they can build, that's the experience I'm aiming for.
And I completely agree about the community. LEGO enthusiasts are incredibly engaged, so earning trust in those communities is probably more valuable than any paid marketing. Thanks for the advice!
Cool! I used to play LEGO many years ago (when I was still young) and I know it 'cause I have this small bag full of mixed LEGO bricks and yes, I have thrown away all the boxes, together with the instructions on how to build the specific model for that set of LEGO, and I must admit I have lost a lot of the bricks from other builds so I would just end up mixing all of them and later on I don't know what to do with them.
So, I guess this would really help me think (only if this was invented many years ago... "just kidding") of what to build with my random/mixed LEGO bricks
I would definitely download this to help me play my LEGO bricks during those times.
the "rediscover what you already own instead of buying more" angle is such a good frame, that's the part that makes it more than a scanner. and honestly a box of mixed bricks nobody can use is a real thing in every house with kids.
Dude this is absolutely BRILLIANT!!!
smart use case for computer vision a lot cleverer than the usual 'take a photo, get info' app since it's solving a real 'what can I actually build' decision problem. Curious how it handles rare/discontinued pieces does it degrade gracefully to 'partial match' suggestions, or does it need a fairly complete standard set to work well?
Thanks! That was exactly the problem I wanted to solve. 😊
Right now, Brickbeam doesn't require you to have a complete set. Instead of looking for exact matches only, it ranks possible builds based on the pieces it detects and can suggest partial matches when you're missing some bricks.
For rare or discontinued pieces, the AI treats them like any other detected part. If a build depends on a rare piece you don't have, it simply lowers that build's match score rather than eliminating all suggestions.
I'm also exploring features like "replaceable parts" and alternative build recommendations, so even incomplete collections can still lead to something fun to build.
It's a good idea but the question is what kind of a moat you can build, especially where LEGO themselves are concerned. One of the big questions here is how difficult it is or is not for LEGO to build a similar feature into their existing software. Right now, it looks like their apps are essentially interactive instruction manuals for sets customers purchase, but this is a tool that could be embedded for customers after they purchase a certain quantity of sets, or reach a certain minimum spend with Lego, etc.
I think the layer that begins to build a moat would be to add a generative AI component that tells people what they could build that is not already in the LEGO catalogue (new, AI generated designs). LEGO would be reticent to incorporate this because it destabilises their branding and sales strategy, and it would give Brickbeam a strong USP
Thanks, this is a really thoughtful point. I agree that the long-term challenge is not just building a LEGO piece recognition tool, but creating something that becomes more valuable over time.
My current thinking is that Brickbeam is not trying to replace LEGO’s ecosystem or compete with their instructions/catalogue. The interesting opportunity is the opposite: helping people unlock more value from the bricks they already own, including the thousands of pieces that are not part of an official set anymore.
I really like your point about generative AI designs. I think that’s one of the directions that could make Brickbeam much more unique — moving from “find existing sets you can build” to “imagine new creations from your own collection.”
There are also some interesting challenges there: making AI-generated builds actually buildable, respecting real-world constraints (available pieces, stability, missing parts), and making the experience fun rather than just generating random models.
Still early days, but feedback like this helps shape where the product should go. Thanks for sharing your perspective!
The interesting challenge is turning a one-time "what can I build with these bricks?" moment into a recurring LEGO companion.
What would convince you that Brickbeam can become part of a family's ongoing LEGO workflow rather than a tool they open once per collection?
Great point. The initial idea was solving the discovery problem, but I agree the bigger opportunity is becoming a LEGO companion over time. We’re thinking about features like keeping a digital inventory, tracking builds, and generating new ideas as collections grow. The goal is not just 'what can I build today?' but 'what can I create next with what I already have?
Appreciate the context.
The shift from a one-time discovery tool to something that supports an ongoing hobby journey is an interesting direction to explore.
The photo-to-inventory step is basically the whole product here, so I'm curious how you handle verification. I work on an app that uses an LLM to extract structured entities from noisy input (place mentions from video transcripts), and the biggest quality lever for us wasn't a better model - it was a deterministic verification layer after it: every candidate gets checked against a ground-truth API before we save it, and anything that doesn't match gets dropped instead of shown. Raw model output was ~80% right; the verified output is what made users actually trust the feature. For bricks the analog might be validating detections against a parts database (part id + color) instead of trusting the vision label directly.
Also curious about unit economics - vision calls on a photo of a 2000-piece pile can't be cheap. Do you cap per-scan cost or downscale before sending? We ended up putting a hard per-job cost guard in front of our pipeline after a few pathological inputs blew past our estimates.
Thanks for the thoughtful feedback! Those are exactly the kinds of challenges we've been working on.
For verification, we don't rely on the raw vision output alone. Every detection goes through additional validation against LEGO part and color data, and we assign confidence scores rather than blindly accepting every prediction. If confidence is too low, we prefer not to identify a piece instead of showing something incorrect. Trust is much more important than squeezing out a few extra detections.
On the cost side, yes—a huge pile can get expensive quickly. We've optimized the pipeline with image preprocessing and batching, and we also have safeguards to prevent unusually expensive scans. There's definitely a balance between accuracy, latency, and cost, and we're still iterating as more people use the app.
Really appreciate you sharing your experience. It's always interesting to hear how similar problems show up in completely different domains.
But the product will live or die on one friction point: can a parent photograph a genuinely messy collection and get a buildable suggestion without manually correcting dozens of detected pieces?
I’d measure one thing ruthlessly: time from first photo to first completed build. If that stays low despite poor lighting, overlapping pieces, and incomplete collections, Brickbeam becomes much more than a clever AI demo. How are you testing it under real household conditions rather than clean photos?
The part about "messy piles of bricks" hits home — I've got a bin in my closet that I keep meaning to do something with but end up just staring at. Is the AI able to recognize bricks just from photos, or do you need to input the pieces manually?
That's a great point, and honestly it's the biggest technical challenge for Brickbeam.
Right now I'm testing with intentionally messy, real-world collections instead of staged photos—mixed bricks, overlapping pieces, different lighting, and partially visible parts. The goal isn't perfect detection; it's getting someone from "box of random bricks" to "I can actually build something" with as little manual correction as possible.
I really like your metric: time from first photo to first completed build. That's probably a better north star than raw detection accuracy, and I'm going to start tracking it more explicitly as I iterate.
I'm sure there's still a long way to go, but that's exactly the problem I'm trying to solve.