Show HN: Buzz Bear – Monitor Reddit with Semantic Search ($0/mo cloud GPU costs)

2 points

5 months ago

BuzzBear monitors Reddit and alerts you when people post about topics you care about. It uses semantic matching, so you can describe what you're looking for in plain English, and it finds relevant posts even when people use completely different words. Example: Search for "CI/CD tools" and catch posts like "our deployment process is a nightmare"

https://buzzbear.ai/ (free tier available)

---

How I avoided cloud GPU costs:

I needed ML inference for text similarity. Cloud GPU pricing is $700-1000/mo for instances that would sit idle most of the time. "Scale to zero" sounds great until you want a generous free tier.

My solution: the RTX 3080 I already own.

The stack (Elixir/Phoenix + Bumblebee):

  - Phoenix app on Fly.io

  - Bumblebee ML node on my home server (RTX 3080)

  - Tailscale + libcluster to connect them

From the code's perspective, they're one Erlang cluster. Where I can call the ML service like a local function. Tailscale tunnels the traffic, libcluster handles discovery. Capacity: ~1M comparisons/hour. I currently need a fraction of that.

Scaling: If I need more, I spin up cloud GPU nodes, connect them to my Tailscale network, and libcluster auto-discovers them. Zero code changes.

Happy to answer questions about any of this!