applied means it has to survive contact with real work. everything here is tested in the field, published when it holds up, and honest about its status.
every answer grounded in real regulatory documents, filtered to the right jurisdiction, with citations. safetylm retrieves from a catalogued corpus of 616 AU/NZ WHS source documents (legislation, regulations, codes of practice) processed into 20,451 metadata-tagged chunks, and runs entirely locally: no API dependency, no per-query cost, full data sovereignty.
most safety dashboards report what already happened. this project catalogues the indicators organisations actually track, then tests which ones genuinely lead: which signals shift before incidents, and which are just noise with a good reputation.
every organisation has a gap between work-as-imagined and work-as-done. this project turns the study of that gap into a repeatable field method: learning teams, friction mapping and procedure-vs-practice comparisons that produce findings you can compare across sites.
long-form psychosocial surveys are rigorous but exhausting, so they run once a year and age immediately. we are exploring a short-form monthly check-in that line leaders can run themselves, designed to track against the long-form instruments rather than replace them.
the full agenda. open questions become projects when we find a way to test them properly, and get retired honestly when we can't.
labs people are part practitioner, part researcher. no open roles right now, but the door is worth knocking on.
labs careers