Just launched the Closed Beta for Link Mesh!
Hey Indie Hackers! After months of coding and refining the architecture, I’m finally opening up Link Mesh for closed beta testing.
The biggest challenge so far was moving the heavy semantic analysis away from local PHP (which would kill any WordPress server) to a dedicated Python/FastAPI backend. Now, the AI can effortlessly analyze sites with 100 to 10K articles without slowing down the user's hosting.
Currently working with a small group of SEO professionals to stress-test the semantic engine. Excited to finally share this journey here!
congrats on the launch! Moving heavy semantic processing away from WordPress/PHP was definitely the right architectural call especially at 10K+ articles scale.
Are you handling async processing with queues (like Celery/Redis) on the FastAPI side? Curious how you're managing large batch analysis.
Thanks. Yes, on the FastAPI side, we use Celery + Redis. Parallel generation. Each anchor task is an independent worker task
Nice that’s a clean setup. Parallelizing anchor tasks makes a lot of sense at that scale.
Are you handling retry logic and task idempotency for long runs? With 10K+ articles I imagine edge cases can stack up quickly.
What I'd be careful about is that solving the scaling problem can sometimes make it feel like the product question is solved too.
The two don't always move together.
Especially when the thing being optimized technically isn't necessarily the thing users end up valuing most.
If you have a WordPress site, I'd love to invite you to the test. I give an LTD and prepaid AI tokens to early users for their feedback
Spot on. It’s easy to fall into the engineering trap where you celebrate building a complex architecture, while ignoring whether it actually solves the user's core problem
I had to solve the scaling issue first, simply because the semantic analysis was physically crashing WordPress on large sites. But you are 100% right, scale doesn't equal product- market fit. Users don't pay for a FastApi backend; they pay for time saved and highly relevant internal links.
That’s exactly why the current closed beta is heavily focused on working directly with SEO professionals. We are stress-testing the "product question" right now to ensure the AI's logic matches human editorial standards
Possibly.
I'd just be careful that validation can sometimes end up answering whether the output is good while leaving a different question unresolved.
Those aren't always the same thing.
That's usually where things start getting interesting.