Alert → one button
Deploy in ~10 minutes
A jellyfish early-warning alert reaches the control room. The operator presses one button and JAZR lowers into each of the four intake lanes.
4 lanes~10 min
JAZR
جَزْر
The plant keeps running.
The sea keeps living.
A one-button, no-harm jellyfish defence for a coastal nuclear plant’s seawater intake. It keeps a bloom off the plant’s screens and returns the animals to the sea alive.
Not a concept. A system.
~99%
of the bloom kept off the plant screens — 100% today → ~1%
76–98%
released alive — two estimates; target ≥95%
−60–90%
crew-hours spent fighting a bloom
~10 min
to deploy in all four intake lanes, one button
≈600
engineering checks
112
automated tests
292
real alert windows replayed
900/900
bells guided in the virtual tank
1.4 pt
browser twin vs Python model
≈0.03%
of plant output in energy
All figures are in simulation, from the team’s engineering models.
How it works
Alert in. Animals out, alive.
An early-warning alert arrives. The operator presses one button. In about ten minutes JAZR is deployed in each of the four intake lanes — and the plant’s own screens are never touched.
Layer 1
Moving-belt angled screen
An 11° screen with a moving belt carries bells along to an apex entrance. Flared entrances take 15–35% of the lane flow, so animals are guided, not pinned.
11° angleApex + flared entrances
Layer 2
Hydro-jet curtain — water only
Any leakers are lifted by a curtain of water jets to a surface skimmer. No air: air trapped under a bell kills jellyfish.
No bubblesSurface skimmer
Dewater + protect
Turtle-excluder grid
Flat-plate dewatering, then a grid with ~15 cm bars. Turtles and big bells take an escape chute, alive. Camera and sonar back it up with a 2-of-3 vote.
~15 cm bars2-of-3 backup
HydroLift
Water-only float chambers
Chambers fill and rise, floating the animals up with the water. Nothing touches them.
Nothing touches the animals
Head tank
Oxygen in, gravity out
A head tank tops up dissolved oxygen — no bubbles — and feeds a 1.6 m gravity pipe through the breakwater.
Dissolved O₂1.6 m pipe
Drift forecast
Near or Far outlet
A 10-day drift forecast picks where to release: Near (5 km out) or Far (~10 km alongshore), so the bloom does not drift straight back.
10-day forecast5 km / ~10 km
Release
Back to the sea, alive
Animals leave through a submerged diffuser at ≤1 m/s — gentle enough to release, not to injure.
≤1 m/s
Bloom Incident OS
Safe by construction
Auto-deploy with operator veto. Hard-wired interlocks. AI is advisory only and air-gapped. Shear-pin relief means any failure leaves today’s plant — never worse.
Operator vetoAir-gapped AIShear-pin relief
Before vs with JAZR
Same bloom. Different morning.
Today, the whole bloom reaches the plant screens. With JAZR, about one in a hundred does — in simulation.
100%
Today
~1%
With JAZR
0.4%best case reaching the plant screens
71% → 6.2%hindcast, 292 real alert windows (earlier design)
Untouchedthe plant’s own screens stay exactly as they are
The prototype
Open it. Run it.
Three working pieces of software, live in your browser — no install.
3D digital twin
The whole intake in real time: belt screens, hydro-jets, excluder, HydroLift, pipe and outlets. Guided camera stops and a before/with split.
Open twin ↗
Bloom Incident OS
The operator’s console: from alert to auto-deploy with operator veto, across all four lanes.
Open dashboard ↗
Live simulation
Open the dashboard and choose the Simulation tab. Run a bloom through the model and watch the outcome update live.
Open simulation ↗Films
See it move.
No harm by design
Rules the machine can’t break.
Every stage was chosen so the animal leaves the way it arrived.
No air under bells
Water-only jets and lifts. Air trapped under a bell kills it.
No nets, no killing
Animals are guided and floated, never caught, crushed or macerated.
No acoustics or electric fields
No sound or electric deterrents — guidance is by flow and structure alone.
Turtle escape
~15 cm excluder bars and an escape chute, with camera + sonar backup.
O₂ kept at 75–99%
Dissolved oxygen in the head tank — no bubbles.
Matched temp & salinity
Animals stay in their own seawater from intake to release.
Outlets away from reefs
Near and Far release points sit away from reefs; the drift forecast picks one for each release.
Evidence & honesty
Tested hard. Honest about the rest.
Test programme
- ≈600Engineering checks on the design.
- 112Automated tests run on the models and software.
- 292Real alert windows replayed. Hindcast on an earlier design: 71% of the bloom reaching the screens → 6.2%.
- 900/900Bells guided to the entrance in the virtual tank.
- 1.4 ptBrowser twin vs Python model — the live demo agrees with the engineering model.
What only real tests can settle
- Survival after the full ride — narrowing 76–98% to a measured number against the ≥95% target.
- How real bells behave on the moving belt and at the apex entrance.
- Hydro-jet lift and skimmer capture in real flow and waves.
- Turtle and large-bell response at the excluder grid.
- Fouling, sand and wear over a full season.
Honest note: there is no physical prototype yet. Every number on this page comes from models and simulation.
Next step
- 1Tank testLive bells, one stage at a time
- 2FlumeFull chain in real flow
- 3One-lane pilotOne intake lane, real bloom season
- 4RolloutAll four lanes
Gallery
Picture it in the water.
JAZR · جَزْر
The plant keeps running.
The sea keeps living.
The team · IIT Delhi – Abu Dhabi
Lakshan Muruganandam
Tarunika Ravikumar
Shouq Alhosani UAE national