How our engineers untangle the hard parts of business software: AI that reviewers and analysts can trust, prescriptions a pharmacist can verify, invoices that check themselves, and apps that never lose data. Each note covers the problem, what we built, and what we would do again.
Every technical claim is checked against the code it describes, and products and clients stay unnamed. For new field notes, follow SeriusTech on LinkedIn
An AI reviewer we helped build flagged 18 errors in a translated book. A linguist found only four right as filed. Here is how we redesigned it for precision and check the real errors survive.
Contractors were retyping numbers finance already had. We built a portal that pre-fills invoices from the HR sheet, checks early ones line by line, and marks them paid on schedule.
Pharmacists can’t tell a genuine prescription from an edited one. Here is how we built prescriptions they verify with a QR scan that holds no patient data and opens a deliberately minimal page.
Many AI research tools let the model write or pick its own citations, unchecked. We built one where every sentence maps to a verbatim quote and copied articles count once.
Users switch phones, restore old backups and log at 1 a.m. on DST weekends. Here is the offline-first design we built for each moment, and where it still falls short.
Grading linguist tests by hand doesn’t scale, and bots grab first-come jobs in seconds. How we built vetting and job claiming that stay fast, fair and defensible.
Most organizations misestimate AI costs, and batch jobs that die halfway can pay twice for finished work. How we make AI batch jobs quotable, capped and resumable.
Cinematic property tours often stutter on mid-range phones and bury the booking form. Here is how we build scroll-driven tours to stay smooth, accessible and measurable.
An agent ticks twelve boxes and reports all tests pass. Here is how we make “done” checkable, block rm -rf with few false alarms and cap unattended runs.
Many UAT sign-off disputes are really evidence problems. Here is how we turn a client handoff into a go/no-go verdict both sides can defend, with screenshots that hide form data.
