Automated HACCP for food manufacturers

The operator describes in natural language what they need and Rela configures the CCPs, the limits, the grace periods. Every deviation is detected in under a second and triggers the corrective action, the root cause record and the BRCGS / FSSC 22000 report.

Rela AI is not yet another HACCP checklist in the cloud. It is the brain of your food plant: it reads your existing PLCs, watches every CCP continuously, detects deviations in sub-second time, opens the corrective action as an assignable work order, logs root cause, places the affected batch on hold and exports the audit report with timestamp, electronic signature and chain of custody. All in a single AI, in Spanish, Italian and English, accessible from WhatsApp and email.

The problem: HACCP on spreadsheets does not scale

Typical case — Italian bakery with 15 million euros in revenue, three white-bread frozen lines for large retail, 42 operators, BRCGS Issue 9 certification current. How does it monitor HACCP today?

• Each shift, two operators write down by hand the temperature of the proofing chambers, ovens and cooling tunnels. Four readings per shift, three shifts a day, seven days a week. Over 80 paper points per day.
• Finished-product freezers have a Carel datalogger with its own app that beeps when something goes out of range — but the app lives on the production manager's phone and nobody else sees it.
• The QA supervisor digitizes the records once a week into an Excel template, fixes by hand whatever looks odd and files it.
• The day before the BRCGS audit, the QA team prints three twelve-centimeter folders, scans them in panic and prays the auditor will not ask for the traceability of a specific batch from last year.

The real cost is not the paper. It is the operational blindness. By the time an operator notices that proofing chamber LIE-A spent 22 minutes at 31°C, the batch has already left the line and, if it reaches a hospital, supermarket or school, the recall costs anywhere from 80,000 to 350,000 euros plus FDA or European authority fines, plus the brand damage. Industry studies place the recall of an industrial bread brand with three-country coverage near 10 million euros.

HACCP on spreadsheets does not scale because detection depends on human presence. Any BRCGS or FSSC 22000 auditor will tell you the same thing: clause 2.10 demands continuous, verifiable monitoring, not discretionary sampling.

What is automated HACCP?

Automated HACCP is the difference between inspecting and monitoring. Inspecting means a person walks over, looks and writes it down. Monitoring means the system watches 24/7 without blinking, and only alerts you when there is a real deviation (not when the chamber door opens for 30 seconds to pull a tray).

Three technical notes to tell automated HACCP apart from "checklist software" in disguise:

• Direct PLC reading — the system reads registers from the oven PLC, the Carel freezer thermostat, the cooling tunnel Modbus TCP. It does not depend on a human typing a temperature into an app.
• Sub-second detection — the latency between a sensor logging an out-of-range temperature and the alert reaching the operator is under one second. Software-only that polls every 5 minutes is not automated HACCP: it is delay dressed up as technology.
• Complete causal chain — deviation detected, affected batch identified, root cause logged, disposition decided (hold, rework, destroy, release), report generated. One single pass, no Excel-pasting.

How Rela AI works with HACCP

flowchart LR
  PLC[PLC on oven freezer pasteurizer] -->|Modbus or OPC UA| VPN[VPN tunnel]
  VPN --> Edge[Edge Gateway or cloud]
  Edge --> Detect[Sub-second deviation detection]
  Detect --> CCP[CCP configured via AI]
  CCP --> Whatsapp[WhatsApp notification]
  Whatsapp --> Action[Corrective action logged]
  Action --> Audit[Audit report BRCGS FSSC 22000 FSMA]

Narrative description, node by node:

1. PLC on oven, freezer, pasteurizer. Rela connects to the equipment you already have — Schneider Modicon, Siemens S7, Allen-Bradley CompactLogix, Beckhoff CX, Carel pCO5, Pego Plus. It reads registers that are already exposed. No firmware changes, no agreement with the PLC vendor.
2. VPN tunnel. Rela deploys an Edge Gateway on a Raspberry Pi (or an industrial mini-PC) that opens a WireGuard tunnel to the cloud. Your internal network stays closed to the outside; only the gateway speaks outward through an outbound port.
3. Edge Gateway or cloud. If connectivity is stable, all processing happens in the cloud. If the plant has recurring drops (rural Italian industrial site, free-trade Colombian zone), the Edge Gateway buffers locally with HiveMQ and Node-RED and syncs when the link returns. Zero data loss.
4. Sub-second deviation detection. Each reading goes through the threshold detection math shared with cold chain. Sub-second from read to evaluation. Transient filter (door opening, defrost cycle) with configurable grace period.
5. CCP configured via AI. Each CCP has its range, grace period, corrective action, responsible person. But it is not configured on a 42-field screen: it is configured in natural language over WhatsApp.
6. WhatsApp notification. When there is a real deviation (not a filtered transient), the responsible person receives a WhatsApp message with full context: which CCP, which asset, value read, band violated, batch in progress, suggested action.
7. Corrective action logged. The operator replies on WhatsApp with what they did. Rela generates the work order with prefix HAC-, status todo, priority urgent, assigned to the responsible person, with back-link to the deviation and the batch.
8. Audit report BRCGS, FSSC 22000, FSMA. Every deviation, every corrective action, every batch disposition, every change to a critical limit lands in _audit_trail. PDF-exportable with timestamp and electronic signature on demand.

Typical CCPs by industry

Real CCPs that Rela monitors today for production clients:

Each one of those CCPs is monitored on paper sheets in most SMEs today. Rela makes them continuous.

Configure CCPs in natural language

The key differentiator of the agentic layer: configuring a CCP is a WhatsApp conversation, not a five-step wizard.

Under the hood: the WhatsApp agent parses intent, validates the asset against _assets, calls POST /api/v1/haccp/ccps, logs the mutation in _audit_trail with actor=luca.bianchi, and starts the metric subscription on the PLC. Latency from message to active CCP: under three seconds.

Product disposition and batch traceability

When there is a confirmed HACCP deviation, silencing the alert is not enough. Regulation requires deciding what happens to the product made during the deviation.

Rela records the disposition as an immutable entry in _haccp_batch_dispositions:

• hold — batch is blocked until QA verifies manually.
• rework — batch goes back to process (when the deviation allows it, e.g. reheat to reach pasteurizer minimum).
• destroy — batch is discarded, with logged reason.
• release — QA reviewed and releases (deviation was bounded and verified harmless).

Each disposition includes: batch_id, corrective_action_id, deviation_id, taken_by, recorded_at, notes. Full chain of custody.

When an inspector asks "what happened to batch B-2026-04-15-A from April 15?", a single query answers:

GET /api/v1/haccp/batches/B-2026-04-15-A/history

And it returns the chain of events: deviation detected at 14:02:45, hold at 14:05:12 by Luca Bianchi, QA verification at 16:30:00, decision destroy at 16:42:18, reason: "sustained exposure for more than 18 minutes at 32°C, Listeria risk ruled out but rotation margin insufficient". Done.

BRCGS, FSSC 22000, FSMA — what Rela covers

Coverage by standard:

We do not sell certification. We sell the tool that makes audit day boring instead of terrifying.

What Rela does NOT do (technical transparency)

Honesty — preserves credibility and reduces buyer anxiety:

• Rela does not sign the HACCP plan. The plan is signed by the human QA owner. Rela monitors, records and reports, but regulatory responsibility stays with the person certified by the authority.
• Rela does not replace physical calibration. Physical sensors (temperature probes, metal detectors, scales) must be calibrated on schedule. Rela can remind, schedule and log the calibration result, but does not calibrate.
• Rela does not decide batch disposition on its own. The system suggests corrective action based on history, but the final call (hold, destroy, rework, release) belongs to the operator or QA. Rela records the decision; it does not replace it.
• Rela does not replace the microbiology lab. Listeria, Salmonella, E. coli testing is external or internal lab work. Rela links results to the batch but does not run the assay.
• Rela does not issue export certificates. If your product needs a sanitary certificate to export, the authority issues it. Rela generates the dossier the authority needs.

What Rela does do: watch 24/7, detect in sub-second time, log everything, open work orders, export reports and give you peaceful sleep between audits.

Documented use cases

• Italian bakery — proofing chamber fleet with Modbus and VPN
• HACCP in industrial bakery
• Cold chain in bakery
• Unified alert inbox

Comparison: software-only HACCP vs Rela

How to deploy automated HACCP in your plant

Seven numbered steps from "install VPN" to "first automated BRCGS audit":

1. Day 1 — Install Edge Gateway. A Raspberry Pi 4 (or industrial fanless equivalent) in the technical room. Network cable to the plant switch. WireGuard tunnel to Rela cloud. Hardware investment: under 200 euros.
2. Day 2 to 3 — Map PLCs. Quick audit with your electrician or systems integrator: which PLC controls which asset, which registers are exposed (Modbus address, OPC UA node ID), which unit. One day's work for a 10 to 30 asset plant.
3. Day 4 to 7 — Bootstrap sources. Rela creates _machine_event_sources. Each source begins reading. The sensor watchdog confirms each PLC delivers readings at the expected frequency.
4. Week 2 — Configure CCPs via WhatsApp. The QA owner writes each CCP from the existing HACCP plan into WhatsApp. Rela creates, validates and activates them. Three to five hours for a plant with 25 CCPs.
5. Week 2 to 3 — Silent learning period. Rela monitors but only reports to an internal QA channel (no work orders yet). This catches misconfigured CCPs, unexpected transients and drifting sensors. Final tuning without operational noise.
6. Week 4 — Production activation. WhatsApp notifications to each CCP owner are activated, automatic work-order generation goes live, batch chain of custody begins recording.
7. Month 2 to 3 — First automated BRCGS audit. The QA team exports the dossier from Rela: HACCP plans, CCPs, deviations of the period, corrective actions, batch dispositions, audit trail of changes. PDF with timestamp and electronic signature. The auditor reviews on a screen, not in six cardboard boxes.

Total time from signature to passed audit: 90 days under typical conditions.

Key benefits

• 100 percent of HACCP deviations detected in under one second vs human shift sampling every 4 hours.
• CCPs configurable in natural language — three hours to configure 25 CCPs over WhatsApp vs three days with a traditional wizard.
• Zero HACCP paper. Shift sheets eliminated. Records are born digital, signed and immutable.
• 70 to 90 percent reduction in BRCGS audit prep time. From three weeks to three days in pilot clients.
• Deviation to work order: under 60 seconds in plants with stable connectivity.
• One-query batch traceability. Full chain of custody, not loose files.
• Typical ROI: 6 to 9 months. The first avoided recall pays for 5 years of Rela.
• One asset graph shared across HACCP, predictive maintenance and OEE. When a HACCP deviation is also a freezer compressor failure symptom, one unified alert — not three teams discovering it separately.