Home / Roadmap

Roadmap

Start narrow. Prove it. Then expand.

A realistic plan that begins with post-stun verification for one farmed species in one controlled context — buildable even with recorded video and simulated process data — and grows through validation toward a deployable platform.

First MVP scope

One species. One context. Real evidence.

Video-based post-stun risk flagging, with manual expert review and basic environmental/process logging — for a single farmed species (e.g. European sea bass, gilthead sea bream, or rainbow trout) in one controlled setting. Deliberately narrow, so it is achievable and verifiable within months.

MVP timeline

What we build at 1, 3, and 6 months.

  1. Month 1–2

    Map & define

    State-of-the-art review, expert/advisor interviews, workflow mapping, target species, and a fail-safe welfare-indicator framework. Output: requirements + first species profile + pilot protocol.

  2. Month 2–4

    Build Lite + Sensor Box prototype

    Offline app + dashboard, low-cost temperature/conductivity logging, batch audit report, and a simulated/recorded-data demo. Output: working MVP device + bill of materials.

  3. Month 4–6

    Semi-field pilot

    Test with a realistic workflow, compare logs to manual observation, measure usability and operator burden, refine alert thresholds, and document cost per site. Output: pilot report.

  4. Month 6–9+

    Optional camera assist & validation

    Collect post-stun video, label visible failure/recovery signs, train a conservative edge model, and design the expert-validation protocol. Output: camera-assist prototype + validation plan.

The 12-month plan at a glance

Overlapping phases — build starts before mapping fully ends; the pilot overlaps build; camera/validation is a later, optional track.

Phase 1 map (mo 1–2) → Phase 2 build (mo 2–4) → Phase 3 semi-field pilot (mo 4–6) → Phase 4 optional camera/validation (mo 6–9+). Matches the milestone list above.
Work packages

How the work is organized.

WP1 — State of the art & requirements

Review standards & literature, pick target species, define indicators and what the system can/cannot claim, advisor review. → SOTA report + requirements.

WP2 — Welfare indicator framework

Specify the Welfare Risk Index across visual, process, environmental, operational signals + expert review. → risk-index specification.

WP3 — Data acquisition prototype

Camera + sensor inputs, timestamp synchronization, batch/event logging, local storage, basic dashboard. → working MVP device.

WP4 — AI risk detection

Detection, tracking, motion/anomaly analysis, multimodal classifier, uncertainty score → human review. → risk-detection prototype.

WP5 — Dashboard & reporting

Live view, alerts, batch history, annotated clips, corrective-action log, exportable reports, advisor review panel. → web application.

WP6 — Pilot validation & business model

Validate anomaly detection, missed-event reduction, usability, audit value, and affordability. → pilot report + business case.

Build maturity

How we de-risk — from simulation to a real facility.

Each stage proves something concrete before the next gets expensive. The software pipeline is buildable now on simulated and recorded data; capital hardware only enters once the value is demonstrated.

EEG / VER equipment appears only at the pilot stage, used for scientific validation with experts — never as a routine production component.
StageFocusHardwareSoftware / AI
0 — Desk & simulationProposal, state-of-the-art, mock dashboard, indicator frameworkLaptop; phone/webcam for demoPython · OpenCV · Streamlit/Next.js · SQLite · Label Studio
1 — Low-cost MVPProve video + sensor capture → simple alertsRaspberry Pi 5 / Jetson Nano · ESP32 · temp/EC/pH/DO/turbidityYOLOv8/v11 · MQTT · TimescaleDB · FastAPI · Grafana · Docker
2 — Research prototypeStronger multimodal AI + expert validationMulti-camera · industrial cam · LED lighting · Jetson Orin NXTracking · pose estimation · time-series ML · fusion · MLflow/DVC
3 — Facility pilotReal workflow, audit trail, usabilityIP67/68 enclosures · PoE cameras · edge server · UPS · tablet · optional EEG/VERSpecies config · out-of-range alerts · audit trail · human-vs-AI comparison
4 — PlatformScale-out product—Edge · Cloud · API · Dataset · Certification Pack

Realistic starting point. The Lite tier (checklist app, stun→kill timer, species SOPs, manual logs) and the data pipeline are buildable today; the camera-assist and validation tiers are research goals, gated on the pilot evidence. See the deployment & cost tiers and the open research questions they answer.

Technology-readiness trajectory

TRL 2–3 today → 4–5 within 12 months via bench + simulated line data + a limited pilot.

Components are mature; integration is the work — a near-term effort, not a decade-long research programme. Projected trajectory; endpoints (TRL 2–3→4–5) are our estimate, intermediate points interpolated.

Scale path to >1 billion fish

Verification is a layer over the 100B+ fish already slaughtered — the >1B "excellent" ceiling is reachable. Log scale (fish/yr).

Pilot 10k–100k → EU finfish retrofit (millions) → species-profile expansion (>1B). Illustrative adoption projection, not measured data; only the pilot figure is committed.