Solutions · Aquaculture
Advice a farm manager can act on at 5am.
AquaMind routes an operational question to the right specialists, convenes an expert panel where the answer is contested, checks it against farm telemetry and safety rules, and escalates to a technical specialist when the evidence will not support a recommendation.
The workflow.
Routing is dynamic because the question is not known in advance. A question about mortality after a storm and a question about feed conversion go to different specialists and hit different gates.
- 01
- Dynamic routing
- 02
- Expert advisory panel
- 03
- Evidence verification
- 04
- Safety gate
- 05
- Telemetry check
- 06
- Technical sign-off
The panel disagrees on purpose
Where nutrition, water chemistry and pathology would give different answers, all three are convened and their disagreement is surfaced to the reader with each position’s supporting evidence. The crew does not average three specialists into one bland paragraph.
Safety gate is deterministic
Dosage ranges, withdrawal periods, species tolerances and prohibited treatments are coded rules evaluated outside the model. A recommendation that violates one is blocked, not softened.
Checked against the pond
Recommendations are validated against current telemetry — dissolved oxygen, temperature, salinity, stocking density, recent feed and treatment history. Advice inconsistent with the readings is flagged before it reaches anyone.
Why abstention matters most here.
A confident wrong answer about water chemistry kills a pond overnight. There is no review queue at 5am and no partner to catch it.
An abstention, as the farm manager sees it
Insufficient evidencequestion "Mortality up 4x in pond 7 overnight after heavy rain.
Treat now or wait?"
I cannot establish a treatment recommendation.
what is missing
· dissolved oxygen reading for pond 7 after 02:00
(last reading 19:40 yesterday, before the rainfall)
· salinity after the freshwater inflow
what would resolve it
· a DO reading now — if below 3.5 mg/L this is
consistent with post-rain stratification, not disease
· a salinity reading at 1m depth
what I can say now
· the mortality pattern is not consistent with the two
pathogens seen in this region this season
· do not treat until DO is ruled out — aeration first
escalated to technical specialist on call · SLA 30 min
That is a feature the farm manager will thank you for. It names the two readings, explains what each would mean, gives the one safe action available now, and puts a human on it.
Escalation is a workflow state
Not a suggestion to call someone. The specialist gets the case with the question, the telemetry, the panel positions and the reason for abstention already assembled.
Aquaculture Operations Pack.
Species & regional knowledge
Knowledge Packs per species and growing region, with dated currency and freshness windows on seasonal guidance.
Water-chemistry policy rules
Deterministic tolerance ranges and interaction rules, versioned, evaluated outside the model.
Telemetry connectors
Read-only connectors to common pond sensor platforms, with staleness on a reading treated as missing evidence.
Evaluation cases from real incidents
Including the ones where the original human call was wrong. Those are the cases worth gating on.
Bilingual output
English and Simplified Chinese, generated from the same evidence set rather than translated afterwards, so both versions cite the same sources.
Mobile-first approval
The technical sign-off is designed to be decidable on a phone, standing at a pond, in daylight.
Intermittent connectivity
Runs are durable. A question asked when the link drops completes when it returns; nothing is lost and nothing is silently retried into a duplicate.