Shellfish cold chain control with AI — from boat to shelf, no jumps in the curve.

The problem

The cold room is under control. The gray zone is not.

In shellfish, a well-managed cold chain in the cold room and in refrigerated transport is the rule. The cold room has a PLC, the truck has a data logger, the tunnel is calibrated. And still, the problem keeps coming back in places almost nobody instruments with the same intensity:

1. The unloading dock. The truck arrives, the forklift driver is on another order, the pallet sits fifteen minutes on an open dock at 20°C. The cold room's probe never notices; the pallet does.
2. The urgent cross-dock. Product that comes in and goes out in the same shift jumps from one truck to the distribution truck without passing through the cold room. The new truck's data logger curve starts from a temperature the old truck had already let rise.
3. The overloaded tunnel. Peak load — Christmas, Easter, export season. The tunnel's nominal capacity is at its limit, and the core of the piece does not reach -18°C within the nominal cycle time. The record says “cycle complete”; the core thermometer does not believe it.
4. The data lives in four systems. The logistics operator sends you a monthly PDF from the data logger. The cold room has its chart. The cross-dock, a notebook. The tunnel, its screen. Nobody has the batch's complete curve on a single screen.

And when a retailer arrives with an alert or the inspector asks for traceability under FSMA 204, someone starts piecing fragments together. They do it against the clock, and in the best case, a dossier comes out. In the worst case, it comes out late.

How it fits the IRIS system

iLEAN does not buy new cold rooms — it stitches the chain at the points where it breaks.

The problem in shellfish is not investing more in cold rooms: it is that the chain has gray zones between the well-instrumented links, and that the batch's curve lives in four systems that do not talk to each other. iLEAN is the putty that fills those joints: wireless probes where there were none, Edge on the dock, integration with the logistics operator's data logger, and an agent that closes the batch dossier.

Edge watches the dock. Connect captures truck, cold room and tunnel. The agent stitches the complete curve and anticipates the jump. The person validates — at-risk batches are inspected before the rest, not after the incident.

The three iLEAN pieces applied to shellfish cold chain control:

• Edge — a terminal with machine vision (CNN) on the dock and in gray zones (cross-dock, tunnel antechamber). It counts door openings, measures how long the pallet waits in a warm zone, identifies which pallet belongs to which batch. Works without a network: if the dock loses WiFi, it keeps recording.
• Connect — captures the truck's data logger (PDF/CSV from the logistics operator, or direct reading if the fleet has modern telematics), the cold room's PLC curve, the tunnel's cycle, and the batch's origin record (fish market, fish farm, import). Graduated capture at every link — from the operator's monthly PDF to the in-line reading.
• Agent — stitches the batch's complete curve (origin → transport → dock → cold room → tunnel → dispatch), anticipates the jump when the truck is running late or the tunnel is close to saturation, and closes the FSMA 204 / IFS / BRC per-batch dossier. It warns the dock lead before the truck lands. It does not reassign trucks on its own — it proposes, the person signs.

See the full IRIS architecture →

Before and after

A chain of isolated links vs. a chain with the iLEAN nervous system

Aspect	Today — isolated links	With iLEAN Edge + Connect + Agent
Gray zone on the dock	No continuous measurement	Wireless probe + vision + waiting alert
Urgent cross-docking	No traceability of the pause	Continuous batch curve, no blind jumps
Overloaded tunnel	Nominal cycle assumed	Nominal capacity + actual load + time cross-checked
Operator's data logger	Monthly PDF, reviewed after the fact	Integrated in real time with the batch curve
FSMA 204 dossier	Manual reconstruction on alert	Ready in minutes, signed by the person
Decision on a suspect batch	“Ship it anyway and we'll see”	Targeted inspection + informed decision

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Fresh + frozen shellfish distributor, 3-5 cold rooms + 1-2 tunnels + dock with cross-dock, exporting to the EU/UK/US.
• Pilot: Edge over the highest-traffic dock + wireless probes in the cross-dock + integration with the logistics operator's data logger + FSMA dossier agent. First value expected within a few weeks — the first early warning of a pallet in a warm zone or a tunnel close to saturation.
• Indicative payback between 4 and 9 months, depending on the historical frequency of detected or suspected breaks and the average cost of a batch recalled/claimed by a retailer.
• Measurable hard lever: ≥ 30% reduction in quality hours spent rebuilding traceability on alert, and a live per-batch dossier instead of an after-the-fact one.

And the quality director's reasonable doubt

“What if the agent gets it wrong and makes us inspect a batch that was fine?” — the AI here does not decide the batch's fate. It proposes a priority inspection of the pallet with the suspect curve; the person validates with their own judgment. The reliability of AI in anchored tasks (reading a curve and comparing it against the range) brought error below 1.5% in the best models^[1]. Even so, iLEAN's three safety rings are there so that no critical decision goes out without a human signature.

^[1] OpenAI paper “Why Language Models Hallucinate”, 2025 — on the reliability of AI in anchored tasks.

Related: dairy cold chain · frozen goods pallet by pallet · fish cold room