themisim.SearchEngine

class themisim.SearchEngine(artifacts_dir, *, index_filename='index.faiss', manifest_filename='manifest.parquet', memmap_filename='vectors.f16.dat', meta_filename='vectors.meta.json', brute_force_max=4000000)

Loads the index + manifest + memmap once; serves many queries.

Construction is moderately expensive (faiss.read_index can take a few seconds on a large index, manifest deserialization a similar amount). Hold one engine instance per process. In Streamlit, wrap with st.cache_resource.

Parameters:

• artifacts_dir (Union[str, Path])

• index_filename (str)

• manifest_filename (str)

• memmap_filename (str)

• meta_filename (str)

• brute_force_max (int)

__init__(artifacts_dir, *, index_filename='index.faiss', manifest_filename='manifest.parquet', memmap_filename='vectors.f16.dat', meta_filename='vectors.meta.json', brute_force_max=4000000)

Parameters:

• artifacts_dir (str | Path)

• index_filename (str)

• manifest_filename (str)

• memmap_filename (str)

• meta_filename (str)

• brute_force_max (int)

Return type:

None

Methods

__init__(artifacts_dir, *[, index_filename, ...])
frame_idx_of(gid)	Row index of this global_id's frame inside its shard.
search(query_vec, *[, k, min_score, ...])	Return Hits ranked best first, temporally diversified.
shard_of(gid)	Absolute path of the .f16.npy shard holding this global_id's frame.
shard_starts()	Global_ids where a new shard/CDF begins, with a terminal total_n sentinel appended.
site_of(gid)	Site code (e.g. 'fsmi') for a global_id.
sites()	All distinct site codes present, sorted.
time_ns_bounds()	(min, max) frame timestamp in ns — the indexed UTC time span.
time_ns_of(gid)	UTC timestamp (ns since the epoch) of this global_id's frame.