Meridian

The problem

Meridian is a semantic search product built on Exa with a shared Python core, FastAPI web API, React app, and Typer CLI.

What I built

Built one shared Python core for Exa-powered semantic search and exposed it three ways: a FastAPI web API, a React app, and a Typer CLI.

How it works

• The Exa Python SDK remains the live provider boundary. Tests use mocked clients and fixtures, so CI never spends quota.
• The default answer provider is extractive and keyless. Set LLM_API_KEY and swap to OpenAICompatibleProvider for model synthesis.
• Local SQLite files under MERIDIAN_HOME hold cache, quota, history, and collections. On Windows the default is %LOCALAPPDATA%\Meridian; no telemetry is sent by default.

Tradeoffs and limits

This project does not currently have a hosted demo. Run and verification instructions live in the repository when available.

What I learned

Added neural, keyword, and auto search, cited answer synthesis, saved collections, a SQLite cache with quota tracking, and Markdown, JSON, and CSV exports.