How Buzzlk works — what we read, how we score it, and why we stay grayscale.
We scrape SL X, count the chaos, then AI roasts it or we pick from our phrase jar. No feelings spared.
We track #SriLanka, #lka, #cricket, #rain, and assorted handles that yell into the void. Fresh chaos drops every 2 hours (Sri Lanka time).
Our LLM (Largely Luck-based Math) reads the timeline and yeets posts into buckets—traffic, rain, cricket, politics, the usual suspects. Whatever doesn't fit goes in "Other," aka the junk drawer where vague takes go to die. Topics are ranked by who complained the loudest. (The AI tried its best. We think.)
We analyze sentiment across all posts from the last 2h. If the crowd is mostly negative (≥40%), positive (≥35%), or neutral, we pick one post that matches that mood—deterministic per window so it stays the same until the next sync. No match? We grab one from the pool. We also show how the mood swung vs 2h ago (saltier, calmer, brighter, or same energy).
Each post gets slapped into three buckets: positive, neutral, and negative. An LLM reads the text (sarcasm, slang, mixed tone—handled) and classifies it. If the LLM is unavailable, we fall back to keyword matching. You'll see the emoticon next to each post on topic detail pages too. It's vibes, not science. (Close enough.)
Pure neutrality—we're not playing favorites. We report on everyone: politicians, cricket fans, rain complainers, traffic martyrs. Add a splash of saffron, blue, or green and suddenly we're "taking a side." So we don't. Grayscale = no faction, no allegiance, no "we're definitely not biased" energy. We're Switzerland for Sri Lankan X. (Okay, fine—it also looks really good.)