Four core capabilities
From finding answers to reading the data — the everyday work of water operations teams.
Water Knowledge Base
Operating procedures, standards, equipment manuals, design documents and textbooks in one place. Formulas and tables are parsed faithfully, and every result traces back to its source.
Knowledge Assistant
For operators and engineers. Answers are grounded in the knowledge base with citations; when no source exists the assistant says so and never invents standards or limits.
Operations Diagnosis
Describe an abnormal condition or alarm and get a structured diagnosis: symptoms, likely causes, checks, actions, and safety/escalation. It asks follow-up questions when information is missing.
Operations Analytics
Upload minute-level pump station or plant data. An analytics engine computes energy, specific energy, pump-combination efficiency and data quality precisely, then the model writes the operations report.
One stack from knowledge to control
AquaBrain is designed alongside BidiTech MPC control and AIPC edge controllers, so insights can flow straight into control optimization.
Why on-premises
Data stays in-house
The model, vector search and knowledge base all run on the utility’s own servers. Operating data and internal documents never go to a public cloud.
Isolation by group and plant
Multi-tenant by design: each group and plant has its own workspace and private knowledge base, while shared standards and expertise can be reused.
Fits existing platforms
Use the branded console, or embed the capabilities into existing smart-water, SCADA and O&M platforms through APIs.
A lift pump station: 18 days of minute-level data, one analysis
About 26,000 minute-level records from five pumps were uploaded; a complete operations report came back within minutes. Key findings:
- 1Identified the most efficient pump combination: about 13% lower specific energy than the everyday combination, yet used less than 4% of the time — a clear scheduling opportunity.
- 2Found two fixed-speed pumps running at a power factor of only 0.71–0.76, with recommendations for reactive compensation or VFD retrofit.
- 3Automatically flagged a wrong frequency scaling factor on one pump’s data acquisition, before it could mislead control or KPIs.
Case anonymized.
Coming soon
BidiTech Acoustic Camera · Inspection to AI
Our own acoustic camera delivers raw multichannel audio, acoustic images and inspection reports; AI classifies faults at spectrum level and pinpoints bearings, seals and more.
Model deployment to the edge
Pump scheduling, dosing and aeration models validated on the platform are deployed to plant-side AIPC edge controllers after approval, with staged rollout and rollback.
AquaBrain FAQ
Common questions on data security, deployment and accuracy.
How is data security ensured?
AquaBrain is deployed on-premises: the model, knowledge base and analytics run on the customer’s own servers, so operating data and internal documents never leave. Groups and plants are isolated by workspace and require login.
What hardware is required?
A single server with two 48 GB professional GPUs runs a 27-billion-parameter model together with the embedding and reranking models, and can be scaled out as usage grows.
How do you prevent hallucinated answers?
Answers are grounded in retrieved knowledge with citations; when nothing is found the assistant says so. Numerical analysis is computed exactly by the analytics engine — the model only interprets and writes.
Can it connect to existing smart-water platforms or SCADA?
Yes. AquaBrain exposes standard APIs so Q&A, diagnosis and analytics can be embedded into existing platforms; operating data can be imported as files or connected to live sources.