Agentic AI · solution
Safety and compliance agents for manufacturing
A safety and compliance agent decides whether the current shift-floor state meets EHS thresholds, and if not, which control to trigger and which log entry to create for the compliance record. It runs against your plant’s statutory rulebook, not a generic template, and every action it takes is auditable.
The decision this agent takes
The agent continuously evaluates observations (vision events, PPE detection, gas / noise / temperature readings, permit-to-work status) against the plant’s EHS rule set. Rule violations trigger one of three actions: raise a corrective control (alarm, interlock, or stop-work signal), notify the safety officer, or write a compliance-log entry against the incident.
Data the agent needs
- Vision events from PPE / zone-intrusion / near-miss cameras.
- Sensor telemetry: gas, noise, temperature, pressure, humidity as relevant to the plant.
- Permit-to-work status from the EHS system.
- Statutory rule set: Factories Act 1948, plant-specific safety SOPs, industry codes (e.g. PSM for chemicals).
- Historical incident and near-miss logs for false-positive tuning.
What it does unsupervised
- Write a timestamped incident entry to the compliance log with the observation and the rule triggered.
- Notify the on-shift safety officer through the plant notification bus.
- Trigger low-risk interlocks (alarms, warning lights) that are pre-authorised in the rule set.
- Reconcile completed corrective actions against open incidents.
Where a human stays in the loop
- Any interlock that stops production is human-authorised. The agent recommends stop, it does not stop the line.
- Rule-set changes go through the safety-committee review, not the agent.
- Statutory-audit-facing log entries are reviewed by the safety officer before submission.
- False-positive tuning is signed off by the safety officer, not silently by the agent.
Where this agent gets it wrong
Honest failure modes and the data-quality prerequisites we insist on before an agent goes live. If your plant does not meet these, the agent runs in advisory-only mode until it does.
- Vision-based PPE detection has known failure modes in poor lighting, from overhead angles, and with atypical PPE variants. Data-quality prereq: reference PPE-detection tests at shift start.
- Sensor drift on gas or noise monitors will cause both false alarms and dangerous misses. Prereq: calibration cadence per the sensor manufacturer’s spec, logged.
- A near-miss the model has never seen will be missed. Weekly review of exceptions surfaces these; new patterns go through explicit rule extension.
- The agent never replaces the statutory audit trail. If the plant’s permit-to-work system is not integrated, the agent falls back to advisory-only and the safety officer maintains the paper trail.
How Ajinkya ships this today
Ajinkya’s existing biometric attendance + Factories Act compliance stack (Ankastra) already handles workforce-side statutory logging for Indian enterprise manufacturers. The vision + sensor safety-agent layer described above is scoped per plant; a productionised out-of-the-box safety agent is [NOT YET SHIPPED].
FAQ
Does this replace the safety officer?
No. It changes their job from continuous observation to exception review and audit. Statutory responsibility stays with the officer.
Can it stop the line if it sees something dangerous?
Low-risk interlocks (alarms, warning lights) yes. Full line stops require human authorisation. This is a policy choice with liability implications, not a technical one.
How does the compliance-log entry integrate with statutory audits?
Log entries are exportable in the formats Indian Factories Act inspectors ask for. Format specifics are tuned per plant and per state during the pilot.
Do you work with existing EHS platforms?
Yes. The agent reads permit-to-work and writes incident records to the platform of record; it does not replace it.
Talk to the engineers who ship this
A 30-minute discovery call gets you a scoped answer on whether this agent fits your plant, what data your team would need to prepare, and what a pilot would look like.
Book a 30-min callReviewed by Amey Kadle, Founder, Ajinkya Technologies. Last reviewed: 2026-08-29.