The problem
Subsea search teams collect large volumes of sensor data, then manually hunt for the few frames that matter.
Turn sonar, radar, ROV video, and streamed subsea data into ranked targets, operator-ready reports, and next-best inspection decisions.
LiquidShell is licensed software for teams with ROVs, AUVs, towfish sonar, crews, vessels, and messy subsea datasets. It does not claim command authority over real equipment.
Subsea search teams collect large volumes of sensor data, then manually hunt for the few frames that matter.
Chat or speak to an agent that routes sonar, video, stream, ranking, and report tasks to typed tools.
Every output carries an engine label, caveat, evidence artifact, and human-review requirement.
Bring sonar imagery, ROV frames, radar-like contacts, or a simulated fixture stream.
The deterministic POC router calls modality-specific tools through typed contracts.
Candidates are scored with confidence, area, rationale, caveats, and source evidence.
Mission reports include artifacts, engine labels, recommended inspection actions, and review notes.
The first wedge is small-to-mid operators with real sonar or ROV data but no enterprise AI perception stack: inspection, search and recovery, environmental survey, aquaculture, and waterfront infrastructure teams.
Turn side-scan passes into candidate regions for operator triage.
Rank likely targets so crews know where to spend scarce inspection time.
Create concise reports from structured evidence without claiming legal conclusions.
Send one old non-sensitive sonar scan, ROV clip, target screenshot, or report. We will process it through the prototype and return a sample evidence packet with target candidates, caveats, and a report draft.