Write a safety checklist for AI Safety Checklist Generator for engineering projects, students input their project design and Claude generates a domain-specific safety checklist based on industry standards (ISO, OSHA, IEEE), flagging potential hazards.

An effective AI safety checklist generator for engineering student projects should convert a project design into a structured review covering hazard identification, risk assessment, compliance checks, controls, mitigation actions, and automatic hazard flags. A practical workflow is: 1) classify the project domain and lifecycle stages, 2) perform a baseline survey and task-based/job-hazard analysis, 3) identify physical, electrical, chemical, mechanical, thermal, radiation, ergonomic, and environmental hazards, 4) estimate risk by severity and likelihood, 5) map applicable requirements from OSHA/industry standards and internal lab/shop rules, 6) recommend controls using the hierarchy of controls, 7) assign PPE only after engineering and administrative controls are considered, and 8) generate action items, approvals, and re-review triggers when the design changes. [15] [15] [6]

For student engineering projects, the generator should produce a checklist with these fields for each hazard scenario: project task or subsystem, hazard source, who may be harmed, initiating event, worst credible consequence, existing safeguards, severity, likelihood, risk rating, applicable requirement, required controls, PPE, verification method, responsible person, and status.

The hazard identification logic should explicitly prompt reviewers to inspect the work area, equipment condition, machine guarding, emergency readiness, ventilation, lighting, noise, electrical exposure, and trip hazards. It should also ask whether shutdown procedures exist, whether users understand storage and organization, and whether the job creates dust, chemical, heat, or excessive-noise hazards. These prompts are well suited to AI extraction because they can be mapped directly from design descriptions, bills of materials, operating steps, and maintenance tasks. [1] [1] [1] [1]

A useful risk model for the generator is a simple 5x5 matrix: Severity 1-5 and Likelihood 1-5, with automatic escalation for high-energy systems, hazardous chemicals, rotating machinery, pressure systems, elevated work, and any task involving students who are not yet qualified. The AI should also require a "worst thing that could happen" statement and an emergency response note for each medium or high risk item.

For risk assessment, the generator should require users to identify materials, quantities, operating conditions, and special circumstances; evaluate toxic, physical, reactive, flammable, explosive, radiation, and biological hazards; consider scale-up risks; and then select controls that keep exposures within limits and prepare for emergencies. A strong prompt set is the five-question method: what are the hazards, what is the worst thing that could happen, what can prevent it, what protects people from it, and what should be done if something goes wrong. [7] [7] [7]

The compliance engine should be domain-specific rather than generic. It should tag each project as, for example, electrical/electronics, machine/prototyping, chemical/lab, construction/civil, robotics, welding/fabrication, or mixed-use, then load the relevant checklist modules and standards references.

• Electrical/electronics module: qualified-person requirement, pre-work inspection, lockout/tagout during servicing, grounding or double insulation for portable tools, GFCI use for temporary power, damaged-cord replacement, hazardous-location suitability, labeling of disconnects, enclosure of energized parts, and approach-distance flags for higher voltage.
• Machine/prototyping module: machine guards, point-of-operation hazards, maintenance isolation, formal LOTO procedures, safe clothing/no loose jewelry, and maintenance training.
• Chemical/lab module: hazard communication program, SDS availability, container labeling, chemical-specific training, spill/fire preparedness, exposure minimization, and review of changing process conditions or scale-up.
• Construction/civil module: housekeeping, first-aid coverage, PPE availability, ladder condition and setup, trip-hazard control, and site emergency access.
• PPE module: documented hazard assessment, engineering/administrative controls considered first, PPE matched to hazard, training on use and limitations, inspection, maintenance, and replacement of damaged PPE.

[2] [5] [14] [3] [4] The generator should recommend controls in hierarchy order: eliminate the hazard where possible, substitute safer materials or lower-energy designs, add engineering controls such as guards, enclosures, interlocks, ventilation, shielding, current limiting, or isolation, then administrative controls such as training, supervision, SOPs, signage, inspections, and restricted access, and finally PPE. It should flag any checklist where PPE is the only control for a serious hazard unless a justification is entered. [7] [4] [6]

For PPE selection, the AI should infer likely needs from the task and hazard source, then require confirmation. Typical mappings include safety glasses with side shields for impact and general lab work, chemical splash goggles for corrosives or pressure/glass hazards, face shields for splash or flying-particle tasks, hearing protection for noise areas, task-specific gloves for chemical, cut, thermal, or electrical hazards, protective footwear for puncture/crush/slip/electrical risks, and appropriate head protection where overhead or electrical hazards exist. [11] [7] [12] [8] [9]

Potential hazard flagging rules should be conservative. The AI should automatically flag designs containing any of the following keywords or features for mandatory human review: exposed conductors, mains voltage, batteries above lab-defined energy thresholds, rotating shafts, belts, gears, pinch points, blades, lasers, pressure vessels, compressed gas, flammables, corrosives, toxic dusts/fumes, welding arcs, elevated work, ladders, confined spaces, autonomous motion, hot surfaces, cryogens, radiation sources, and any bypass of guards or interlocks.

Electrical hazard flags should specifically include exposed live wiring, servicing without isolation, missing grounding, damaged cords, lack of GFCI protection for temporary power, use in wet areas without suitable equipment, unlabeled disconnects, uncovered openings in electrical boxes, and work near high voltage. For machinery, flags should include missing guards, need to bypass guards during maintenance, body entry into danger zones, and absence of formal lockout/tagout procedures. [1] [2] [5] [10] [14]

A strong output format for each generated checklist item is: Hazard: [description] Trigger/task: [when it occurs] Potential harm: [injury/loss] Risk rating: [severity x likelihood] Applicable requirement: [OSHA/ISO/IEEE/internal rule] Required controls: [engineering, administrative, PPE] Mitigation actions: [specific steps, owner, due date] Verification: [inspection/test/training/document] Escalation flag: [none/review/stop-work] This structure works well for student projects because it is auditable, teachable, and easy for faculty or lab managers to review.

If you are implementing the generator, use retrieval-augmented prompts with a controlled taxonomy, not free-form generation alone. Require the model to cite the matched checklist rule, explain why it applies to the design, and state assumptions. Add confidence scoring and route low-confidence or high-risk outputs to a qualified reviewer.

• Minimum inputs: project description, subsystem list, schematics/CAD/BOM, materials and chemicals, operating voltages/pressures/temperatures, fabrication methods, maintenance tasks, test procedures, workspace type, user skill level, and emergency resources.
• Minimum outputs: domain-specific checklist, prioritized hazard register, compliance gaps, recommended controls, PPE matrix, training requirements, inspection/test checklist, and stop-work flags.
• Governance controls: version control for standards mappings, human approval for high-risk items, logging of model decisions, periodic review against incidents/near misses, and clear disclaimer that the tool supports but does not replace competent safety review.

[15] [15] [13]