---
title: WaterScopeAI Search
emoji: 🔎
colorFrom: blue
colorTo: green
sdk: docker
app_port: 7860
pinned: false
---

# WaterScopeAI Search System

CPU-based hybrid publication search, retrieval, and MCP backend for WaterScope-AI.

This Space is intentionally separated from the GPU model Space (`EnvironmentalAI/WaterScopeAI`). Large corpus assets and the canonical prebuilt Qdrant index will be connected from a Hugging Face Dataset repository in the next deployment phase.

## Initial deployment test

- `/` — service information
- `/api/health` — backend/resource health
- `/api/search?q=activated+sludge&top_k=10` — publication search

The first package includes `papers.csv` and the paper manifest so BM25 search can be tested before the prebuilt Qdrant index and document assets are connected. `METADATA_AUTO_REBUILD=0` prevents the CPU Space from rebuilding the vector index automatically; if no prebuilt Qdrant index is present, search falls back to BM25.