LuisSearch

LuisSearch is a real, independently-built search engine covering both general web content and a dedicated image index, alongside AI-assisted features layered on top of classic keyword search.
Features
- Web search -- a from-scratch crawler and BM25-style ranking index over crawled pages, with a Host Name Boost that surfaces official domains (e.g. docs.docker.com for a 'docker' query) higher in results
- Image search -- a separate crawler indexes
<img>alt text from crawled pages into its own database, with thousands of images indexed - AI reranking -- a local LLM reranks the top candidates from the keyword index for queries longer than a couple of words, falling back to plain keyword order if the AI call times out
- Public API -- documented endpoints for search, image search, and account/API-key management, with tiered rate limits (none/free/account/plus)
- Nexus -- a real-time chat feature built into the same codebase, using long-polling rather than WebSockets or Server-Sent Events (the Cloudflare tunnel LuisSearch runs behind buffers SSE, so long-polling was used instead)
- AI-generated sites -- users can describe a small site or app and have it generated, deployed, and hosted directly from the platform
Architecture
LuisSearch runs as a single Python stdlib http.server-based backend (no framework), backed by SQLite for its indexes, accounts, and logs. The public-facing domain is a Cloudflare Pages deployment whose _worker.js proxies API traffic through to the real backend, which itself runs behind a cloudflared tunnel rather than being hosted directly on Cloudflare's own compute.
History
The image search feature and public API were both added as later milestones after the core web search was already live, expanding LuisSearch from a plain keyword search tool into a small multi-purpose platform combining search, chat, and AI-assisted site generation.