joshua builds Leviathan, a record indexer for agents working over large datasets. Their public background includes high-frequency trading infrastructure and current work on intelligent manufacturing.[1][2][3]
Start with Leviathan →
X · joshuagunnn Career
Their GitHub biography describes earlier high-frequency trading work and a move into intelligent manufacturing. The linked personal site gives the same account. Leviathan addresses a related information problem: making relevant records available within an agent’s context budget without treating an entire operational history as one prompt.[2][3]
Contributions
Retrieve a few records instead of rereading the dataset
Leviathan converts structured exports into a ranked full-text index stored in SQLite. An agent can search records and receive short result cards with identifiers instead of reading an entire table. Field mappings define titles, text, groups and dates; filters and group resolution make the scope of a query explicit.[4]
Keep a changing record collection current
The tool supports indexing files, database CLI output and multiple datasets, with upserts and deletion for later changes. An optional MCP interface exposes read-only search and retrieval operations. Its README performance example uses a synthetic maintenance log, so the reported accuracy and token savings depend on that dataset and comparison method.[4]
Tools, writing & talks
Tools
Start here
Leviathan
github.comA structured-record index with ranked search and compact, identified result cards for agents.
Start with the sample ticket dataset and field mapping. It needs the binary or a Rust build environment; compare retrieved records with your own expected answers before relying on the supplied benchmark.
Creator and repository owner[1][2][4]Original source