Skip to content
Lubo Bali
All projects

AI agentsWeather · Government

SkyIndex-AI

Search that finds the answer even when the words don't match

30 minbetween refreshes; only changed text is processed again

See the code on GitHub (opens in a new tab)

Proof you can check:Code on GitHub

AI agents

Finds the answer even when the words don't match

  • “will it rain tomorrow morning”Forecast: a slight chance of showers and thunderstorms0.627
  • “dangerous heat, what should I do”Heat Advisory: reschedule strenuous activities0.664
  • “flash flood risk this weekend”Flash Flood Warning, El Paso, TX0.517

Lubo Bali · lubobali.com

No. 11Live National Weather Service text

The problem

Important answers are buried in free text, and keyword search misses them when the writer used other words: a forecast says "showers", you typed "rain".

What I built

A pipeline that harvests live weather alerts and forecast discussions, turns them into vectors in Postgres (pgvector) and serves a search where meaning matters, not exact words. An AI can write a short answer from the passages it found.

Tech & why

Lakebase Postgres with pgvector keeps search and data in one database; each text is processed once and only again when it changes, so refreshes stay cheap.

  • Python
  • pgvector
  • Lakebase
  • Databricks

The result

"Will it rain tomorrow morning" finds "a slight chance of showers and thunderstorms", which keyword search returns nothing for. 221 tests, refreshed every 30 minutes.

How this agent is built

Every AI agent has the same five parts. Here is what each one is in SkyIndex-AI, so you can judge any agent the same way.

  1. Modelall-MiniLM-L6-v2 turns text into vectors; an AI model can write a short answer from the passages found.
  2. HarnessHarvest, chunk, embed, store; a scheduled refresh every 30 minutes re-embeds only text that changed.
  3. GuardrailsAnswers come only from the retrieved passages, and say so when the passages don't hold the answer.
  4. KnowledgeLive National Weather Service alerts and forecast discussions in Lakebase.
  5. ToolsA search API over pgvector (HNSW index).

Similar projects

Have a problem like this? Get your free AI blueprint in about a minute.

Tell me your problem. I'll find a way to build the solution.

Available now

Book a free 30-min call (opens Calendly in a new tab)

No obligation. Pick a time that suits you.

Prefer email?data@lubobali.com