Semantic emoji search with on-device bge-small beats LLM at $0 per search
maxleiter · x · 2026-09-27
Responding to the buzz around Linear adding an LLM ("Jev") to its emoji picker, Max Leiter argues developers are treating LLMs as a hammer for every classification task. Two years ago he built semantic emoji search with no LLM: bge-small-en-v1.5 (37MB) runs on-device in the browser, downloads once, and every search costs $0 — faster, cheaper, and occasionally better than the LLM route. His takeaway: use LLMs only when you actually need reasoning behind the decision; embeddings (plus pgvector for the API variant) suffice for tasks like emoji search. The post includes a live demo comparing on-device, Embeddings API, and Jev costs.
More from coding & agent
- He replaced a $5,000/month ghostwriter with 11 Claude Skills and earned $47,000 from one post — aryanXmahajan · 2026-11-11
- Refero Styles ships 2,000+ DESIGN.md files to give Claude Code real design taste — bbssppllvv · 2026-11-11
- Agent circuit breaker: openclaw shorts bot halts itself after 3 failed experiments — Rich_Standard_1724 · 2026-11-11
- The next AI vendor lock-in won't be the model API, it'll be the harness — Arindam_1729 · 2026-11-11
- Stop burning tokens: a Claude Code orchestrator prompt that routes tasks to cheaper models — TawohAwa · 2026-11-11
- Amazon paper: multi-agent setups burn 1.4-4.7x tokens for small gains; single agent is the default — rohanpaul_ai · 2026-11-11