Fish abundance
Counts by species group, around the clock — with full clips saved whenever something interesting swims through the frame.
Species groups24/7Event clips
- Reef restoration
- Edge AI
- Open hardware
- 3D reef twins
- Coral settlement
- Mitigation compliance
- Bleaching alerts
- Photogrammetry
- Biodiversity credits
- Printable instruments
The problem
The ocean restoration economy is flying blind — you cannot finance, insure, or manage what you cannot measure.
NOAA’s Mission: Iconic Reefs is a $100M-scale rebuild of the Florida Keys. Florida law requires developers to build and monitor mitigation habitat. Reef insurance and biodiversity credits are emerging worldwide. And underneath nearly all of it, verification is still divers, clipboards, and quadrats — a few days a year.
01
Snapshots, not signals
A dive-survey day costs $1,500–$3,000, covers a tiny area, and happens a few times a year. The 2023 Florida heatwave killed coral nurseries in days — annual monitoring never saw it coming.
02
Priced for oil & gas
Commercial monitoring stations run $10K–$50K+, built for offshore industry — not reef NGOs, mitigation contractors, or Caribbean marine parks. The result: most restoration sites have no continuous monitoring at all.
03
Unverifiable claims
Funders can't compare providers. Compliance reports are sparse and subjective. Ecosystem-service and biodiversity-credit deals stall because buyers don't trust the numbers.
The system
Three pieces. One proof layer.
ReefProof turns any restoration site into a continuously verified, fundable, insurable outcome — with instruments anyone can print. We don't build reefs; we're the missing layer under everyone who does.
01v0 · printing
WITNESS POD
Printable underwater AI camera
A 3D-printed housing around a 4-inch acrylic dome — or printed end-caps on standard PVC pipe — carrying an NVIDIA Jetson running open vision models at the edge. It counts fish, detects settlement, and measures cover and bleaching on the device, storing weeks of metrics per SD-card swap.
$500
BOM
100×
Cheaper
1–10 m
Depth (v1)
02v0 · pipeline
REEF TWIN
Millimeter-scale 3D change tracking
A service-dive sweep with an action camera becomes a full 3D reconstruction on any RTX machine — CUDA photogrammetry and Gaussian splatting. Twins are diffed visit-to-visit to quantify growth, erosion, and structural complexity: the metrics scientists already trust, produced 10× cheaper.
mm
Resolution
3DGS
Pipeline
10×
Cheaper
03v0 · spec
PROOF REPORT
Auditable restoration outcomes
Survival rates, cover change, biodiversity uplift, bleaching alerts within 24 hours — timestamped, standardized, and re-runnable by any auditor, because the models and hardware are open. Closed-box metrics don't sell credits; auditable ones do.
< 24 h
Alerts
Open
Models
Per-site
SaaS
Honest scope — v1 targets the 1–10 m sites where restoration actually happens. We're not eDNA, not aquaculture vision, and not a reef builder — we're the verification layer for the people who are.
What it measures
The numbers a reef lives or dies by.
Every metric is produced against a printed calibration tile of exactly known geometry — so growth is measured in millimeters, not vibes, and every deployment on Earth is directly comparable.
Cover & bleaching
Coral cover and bleaching fraction estimated continuously on-device, with alerts inside 24 hours when whitening spikes.
SegmentationAlertsOn-device
Juvenile settlement
The calibration tile doubles as settlement substrate — new coral and oyster recruits are detected right on the target.
RecruitsPrinted tile
Structural complexity
Rugosity and habitat structure from 3D twin diffs — among the strongest predictors of reef biodiversity.
Rugosity3D twin
Growth & erosion
Colony-scale change measured in millimeters between visits, against printed targets of exactly known geometry.
mm scaleTwin diffs
Threat flags
Predator outbreaks, algae overgrowth, and gear damage flagged early enough for crews to actually respond.
PredatorsAlgaeEarly warning
How it works
From printer to proof in five steps.
The whole loop rides on dives restoration crews already make — ReefProof just makes every one of them count as evidence.
Current step
01/ 05
Print & deploy
Print bed → seafloor
Then the loop repeats — every service dive extends the record, and the record is the product.
Step 01Print bed → seafloor
Print & deploy
Print the pod and calibration tile on any hobby printer — PETG with an epoxy seal, or the PVC-pipe variant for real water. One dive sets both beside the corals being restored.
PETG / ASA4″ acrylic domeCERN-OHL-P
Step 02Edge AI active
Watch
The pod watches around the clock. Edge AI counts fish, cover, and settlement on the device — no uplink needed underwater. Weeks of metrics per SD-card swap, or live through a surface buoy.
Jetson Orin NanoTensorRT24/7
Step 03Diver in water
Sweep
On each service dive — the same dives restoration crews already make — a swimmer or a cheap ROV sweeps the site with an action camera. Sixty seconds of video is enough.
Action camROV optional
Step 04Twin rendering
Reconstruct
The sweep becomes a millimeter-scale 3D twin on any RTX machine — CUDA photogrammetry and Gaussian splatting — and the pipeline diffs it against every previous visit.
CUDAGaussian splattingmm diffs
Step 05Report signed
Prove
A timestamped Proof of Restoration report: survival, cover change, biodiversity uplift. Auditable by anyone, because the models and the instrument are open.
AuditableOpen modelsCredit-grade
$0
Pod bill of materials
0×
Cheaper than incumbents
>0%
Sites unmonitored today
<0 h
Bleaching alert window
Open by design
The openness is the moat.
Auditors can re-run our models on the raw imagery. A marine park in the Caribbean can print the same instrument we deploy in Florida. Standards win by adoption — repositories ship with our first public release this fall.
- HW
Witness Pod housing — full-print and PVC variants, dome clamp, Jetson tray
CERN-OHL-P - ML
Reef detection models — YOLO + SAM 2 fine-tunes for fish, cover, and settlement
Weights - SPEC
Calibration & settlement tile — printable ArUco target of exactly known geometry
Standard - CODE
Reef Twin pipeline — CUDA photogrammetry / Gaussian splatting + change diffing
Pipeline - DATA
Ocean restoration economy dataset — 188 projects and 42 winners, 2014–2025
Dataset
github.com/hamilton-marine — public September 2026
The crew
Students, scientists, divers.
Hamilton Marine Biosciences is a student-led research group of ML engineers, marine scientists, and divers. We build open instruments because reef NGOs shouldn’t need oil-and-gas budgets to know whether their corals are alive.
We built this broke — on hobby printers, borrowed GPUs, and PVC pipe. That’s the point: if we can print it, so can a marine park anywhere on Earth.
ML & CVOpen hardwareMarine bioDive ops3D printing
Contact
Put restorationon the record.
Pilots, partnerships, permits, press — we answer every serious message. Find us on the road to Ocean Exchange 2026 in Fort Lauderdale this October.
research@hamiltonmarine.aiOcean Exchange 2026 · Fort Lauderdale · Oct 25–27