01
Profiles and grade lines
Static and dynamic grade lines, pressures, velocity, head loss and pipe pressure class at every point, with recommended air valves and drains.
versortech · SAASim · water supply
From street map
to verified network.
AI explores.
The engine decides.
SAASim is versortech's water supply system simulator: profiles, branched and looped networks, PRVs and boosters planned by a , with a report ready for the engineer's review and sign-off.
Version 4.0 stable for field testing · no public access
Workflow vision · SAASim
- 01
The territory
Map · streets · elevations · population
- 02
AI organizes
Request → assumptions confirmed by the engineer
- 03
The engine calculates and judges
NBR 12218 · NBR 12211 · EPANET 2.2 GGA
- 04
AI explores alternatives
Layout · pipe sizes · PRVs · zones, each checked by the engine
- 05
The engineer decides
Comparison · report · sign-off
The workflow we are aiming for. Today the engine, planners and reports already work; AI explores alternatives in experimental mode.
Opportunity
One population change.
The whole network to redo.
Ground elevations, street layout, population forecast, head losses, PRVs, pressure class, air valves, drains and the report. Every revision repeats the cycle, usually across spreadsheets, CAD and separate simulators. SAASim brings these steps into one engine and gives time back to comparing alternatives and deciding.
Explore the idea
Change an input.
Watch the hydraulic grade respond.
A 3 km line fed by a reservoir. Change the population, the pipe size and the PRV, and follow the static and dynamic grade lines, pressures and verdict, as in SAASim.
An illustrative simulation with simplified calculations (Hazen-Williams, C = 140, no minor losses, 200 L/person·day, K1 = 1.2 and K2 = 1.5). This is not the SAASim engine, and the results must not be used for engineering design.
SAASim · Line profile L-01Interactive simulation · simplified calculation
Design flow: 16.7 L/s
- Ground
- Static grade line
- Dynamic grade line
- Reservoir
In the real SAASim
The planner picks each PRV's position and setting for the fewest devices, checks cavitation and pipe pressure class at every point, recommends air valves and drains and, when PRVs are not enough, proposes boosters or larger pipes, explaining why.
Checks
| Minimum dynamic pressure | 15.0 m | ≥ 10 m | passes |
|---|---|---|---|
| Maximum static pressure | 51.0 m | ≤ 40 m | fails |
| Unit head loss | 4.86 m/km | ≤ 10 m/km | passes |
| Maximum velocity | 0.87 m/s | ≤ 1.5 m/s | passes |
Does not meet all criteria checked in this simulation.
P_din ≥ 10 · P_est ≤ 40 · J ≤ 10 m/km · v ≤ 1.5 m/s
What sets us apart
Standards studied. Methods coded. Results checked against EPANET.
Every SAASim formula comes from a standard or a published method, with its source recorded, explicit units and a verification status. Hazen-Williams in the EPANET 2.2 form, Todini's GGA for looped networks, dynamic programming to place PRVs and boosters, and the method of characteristics for transients.
The engine is the single source of numbers: the interface, the planners, the reports and the AI all read from it. Nothing the AI proposes becomes a solution without going through it.
- published verification cases, with source and tolerance
- 31
- published verification cases, with source and tolerance
- automatic comparison for profiles, trees and loops
- EPANET 2.2
- automatic comparison for profiles, trees and loops
- automated tests in version 4.0
- 3,268
- automated tests in version 4.0
- criteria profiles: ABNT and utilities
- 5
- criteria profiles: ABNT and utilities
Validation process
Test, compare and review.
- 01Worked examples from textbooks, courses, manufacturer manuals and EPANET 2.2 cases run on every engine change.
- 02Every code value has a cited source, scope and status; anything still to be checked against the original is logged as pending.
- 03The report says "meets/does not meet the checked criteria", with no seal: responsibility stays with the engineer of record.
Digital elevation models (SRTM, Copernicus) are for preliminary studies. For detailed design, SAASim takes surveyed elevations.
Standards and methods
- NBR 12218:2017Distribution networks
- NBR 12211:1992Design studies and demand
- NBR 12215-1:2017Transmission mains
- NBR 12214:2020Pumping
- EPANET 2.2Hazen-Williams, GGA, valves
What SAASim already does
From a transmission main profile to a neighborhood loop.
All of this works without AI, in the browser, on the same engine.
02
Branched and looped networks
Continuity on trees and GGA on loops, with multiple sources, closed valves, and PRVs and boosters inside pipes.
03
Automatic PRVs and boosters
The planner finds the fewest devices and the lowest feasible settings, with cavitation checked. Sectorization and elevation-band zoning as alternatives.
04
Size and correct
Pipe size per link by lowest reference cost, with engineer locks, and a correction plan showing the before and after of each proposal.
05
Territory and data
Streets suggested from an area, single or dual mains, SRTM or surveyed elevations, population fitted to census data and EPANET (.inp) import.
06
Deliverables
Report with calculation record, critical path profiles, quantities, DXF in UTM, KML, GeoJSON and spreadsheets. The project keeps its revisions.
The AI workflow we want
A conversation about the territory. Every number from the engine.
Today SAASim's AI already explores pipe, PRV and layout alternatives within the engineer's parameters, in experimental mode. We want to reach a continuous flow, from request to delivery. Step through the stages and compare the vision with what exists today.
SAASim · Model neighborhood · proposed workflowScripted conversation · product vision
01 · The engineer describes the project
Request
Where we want to be
The engineer marks the area on the map and describes the request in plain language: horizon, utility, reservoir. AI drafts the assumption sheet and asks only for what is missing.
TodayVision
The engineer fills in the inspector (criteria, demand, source pressures). AI does not take part in this stage.
Conversation · scripted
Engineer
Distribution network for this subdivision, 20-year horizon, local utility criteria, reservoir at the high point.
SAASim (vision)
I drafted the assumption sheet. Three inputs are missing: per capita demand, the reservoir's minimum level, and whether main roads get single or dual mains.
An illustrative dialogue of the workflow we are aiming for. Nothing is sent, calculated or saved.
- In use
- Experimental
- In development
- Vision
Principles
Rules that hold in every phase.
AI widens the exploration. Professional responsibility stays where it has always been.
01
One engine
Every number comes from the deterministic engine. AI never writes an elevation, criterion, result or verdict.
02
Engineer inputs are law
Criteria, demand, pressures, elevations, catalog and locks change only by the engineer's hand.
03
Propose, never apply
Applying belongs to the engineer named in the project, by layer or row, with Undo and a record.
04
No "optimal", no seal
AI presents checked alternatives next to the deterministic solution. "Nothing better" is an expected result.
05
Minimal data to the provider
Only numbers and identifiers, geometry shifted and rotated, with consent. No names, addresses or coordinates.
06
Reproducible without AI
The recipe of an applied alternative reproduces the design and the verdict with the same engine, without calling AI.
Solutions under study
What research points to and how we bring it to SAASim.
Ideas from recent work on AI for water networks and engineering, filtered through our principles. These are research directions, not finished features.
Query
Ask the network
Plain-language questions ("where does pressure drop at peak?", "which link limits the flow?") become engine queries. Answers use only engine numbers and are checked before delivery.
Governance
The engineer as the release gate
In chip formal verification, AI agents propose, deterministic tools prove and experts release. It is the same arrangement as SAASim: AI explores, the engine judges, the engineer signs.
Exploration
A portfolio of alternatives, not an "optimum"
Layout and pipe sizes optimized together, compared by device count, reference cost and resilience (Todini index, N-1 criticality), without collapsing everything into one number.
Heuristics
Heuristics discovered offline
Language models generate design heuristics that are tested on a corpus of networks with known optima. The winner enters as reviewed, tested code; no generated code runs in the product.
Evidence
Evidence before promotion
AI leaves "experimental" only when it beats a deterministic local search with the same budget, across a corpus of networks, with no regression in the worst case.
Review
AI for the people who review designs
The utility reviewer mode already opens the project file and recalculates with the engine. The vision is AI pointing out what to check first, always with the number and the link it came from.
References point the way; every idea we adopt goes through the same engine tests and review.
Roadmap
Where we are and where we are going.
The engine and the deterministic tools come first; AI grows on top of them, phase by phase, with measurement.
01
Today · version 4.0
Stable for field testing
Profiles and mains · In use
Grade lines, pressure class, estimated transients, air valves and drains.
Branched and looped networks · In use
EPANET 2.2 GGA, multiple sources, valves and source types.
Planners · In use
Automatic PRVs and boosters, sectorization, pipe sizing and correction plans.
Territory and delivery · In use
Streets from an area, elevations, census, .inp, reports, DXF, KML and GeoJSON.
AI: explanatory text · In use
Explanations of engine results, checked number by number.
AI: explore alternatives · In development
Pipes, PRVs and tree layouts, in experimental mode.
02
Next steps
After version 4.0
Fire flow · Planned
Fire flows at nodes and network checks for that scenario.
Losses and leakage · Planned
Pressure-dependent leakage (FAVAD) in demand and results.
Elevation uncertainty · Planned
Terrain model uncertainty carried into the verdict.
Trenching and demand allocation · Planned
Excavation cost and demand spread over IBGE census tracts.
AI on looped networks · In development
Closing and opening loops, swapping streets, zone boundaries, with robustness testing.
AI evidence gate · In development
Evaluation against the deterministic local search to leave experimental mode.
03
Vision
Where we want to take the platform
Conversational workflow · Vision
From request to delivery in one conversation, every number from the engine.
Ask the network · Vision
Plain-language queries answered by the engine.
Asset records and surveys · Vision
The utility's existing network and field surveys as inputs.
Offline heuristics · Vision
Heuristics discovered with AI, adopted as reviewed code.
versortech platform · Vision
SAASim next to fablecalc, with the same engine, review and integration standard.
Roadmap targets are subject to change and validation. "In use" means version 4.0 features, in field testing and with no public access.
Who is building it
Meet the team
Filipe Carboni Fim
Founder · Civil engineer
João A. de Barros
Co-founder · Software and product
Vinicius Carboni Fim
Originator of SAASim · Civil engineer
Build with the people who design and operate
Let's test SAASim on real networks.
We want to hear from designers, utilities, municipalities, educators and water researchers. Our next step is to bring version 4.0 to field tests with people who design and review water networks.
These links open a WhatsApp conversation.
SAASim interest list
Leave your contact details to get news and invitations to SAASim field tests.
The essentials about SAASim.
Is SAASim available yet?
Version 4.0 is stable for field testing with the team and partners, with no public access yet. Profiles, branched and looped networks, planners and reports already work.
Does the AI design the network?
No. AI proposes alternatives and drafts explanations. The deterministic engine calculates and judges each alternative, next to the solution without AI. The engineer decides what to apply and signs the design.
Does it work without AI?
Yes. All simulations, code checks, planners and reports work without AI. It is optional and asks for consent before every session.
What data goes to the AI provider?
Only numbers and identifiers, with the geometry shifted and rotated. Names, addresses, coordinates and project identification are not sent.
How does SAASim relate to EPANET?
The engine is our own and uses the same formulations as EPANET 2.2, with automatic comparison on every change. SAASim imports and exports .inp files.