wtf is nexus-lunar?

nexus-lunar, branded Lunar Nexus, is a Python tool that looks for hidden connections between three types of space related text and data: NASA articles, lunar sensor readings, and Apollo astronaut transcripts. It treats these documents as a kind of living data lake, one that keeps ingesting new material and looking for surprising links rather than just storing files for someone to search later. The core process runs in phases. First, ingestion turns each document into a 384 dimensional vector using a sentence embedding model called all-MiniLM-L6-v2, so documents about similar topics end up pointing in similar directions even when they use different words. These vectors are stored in SQLite and in Qdrant, a vector database. Next, a script called pulsar scans the vectors and looks for pairs of documents that are close together, using a measurement called cosine similarity, with a distance threshold of 0.65. Another script, harmonie, looks for three connected pairs that form a triangle across all three document types, an article, a sensor reading, and a transcript, since the project treats a three way match as much more meaningful than a single coincidental pair. Later phases add a 72 hour human review step and a script that generates hypotheses and feeds them back into the data lake so future cycles can build on past discoveries. Right now the project is a manual proof of concept: a person runs the ingestion and detection scripts by hand rather than the system running continuously on its own, though the README describes automatic, always on ingestion as a future goal. To run it you need a Mac with Apple Silicon, Python 3.12 or newer, and Docker, since Qdrant runs inside a Docker container. Setup involves creating a virtual environment, installing Python dependencies, starting the Qdrant container, and creating a local SQLite database with a defined schema. The project is released under the MIT license and has not yet attracted any stars.