BuildProof Smart Inspection Mobile App

BuildProof Smart Inspection helps assessors keep evidence, judgement, and reporting together on site.

Replacing paper in the field. Government assessors inspecting construction sites faced an inefficient workflow split across printed sheets, personal phone cameras, handwritten notes, and manual office-based report compilation. BuildProof Smart Inspection unifies these steps into a field-ready, offline-first mobile tool.

BuildProof Smart Inspection mobile app montage — The app keeps daily tasks, checklist decisions, defect evidence, and report review in one field workflow.

This was part of a larger programme for a Singapore government construction authority . The companion BuildProof Applicant Portal is covered separately.

Context & Problem

Reconstructed evidence risks. Field inspections required high precision. Splicing photo records across personal devices and compiling reports days later created major data compliance risks when construction site conditions changed.

Split evidence

Defects needed photos, remarks, classification, and clause relationships, but those records lived in separate tools.

Delayed reporting

Formal reports were compiled later, creating risk when site conditions changed or defects had to be reconstructed from memory.

Unreliable signal

Construction sites could not guarantee connectivity, so a digital tool had to treat offline work as a normal state.

Core Strategic Constraints:

Offline operation: The application must function without any cellular connectivity on deep-level construction sites.
Auditable review trail: Retain high assessor control, enabling manual edits and clawback overrides on any AI-generated finding.
Many-to-many references: Support linking a single physical defect to multiple compliance checklist clauses.
Immediate compilation: Generate and authorise an official report directly on the device before leaving the site.

The Process: Behind the Scenes

Shadowing assessors in the mud. I spent several weeks shadowing government assessors on-site. We clustered recurring pain points around manual transcription and data loss, identifying that a field tool must operate like a checklist utility first and a camera second.

Interactive Form Prototyping. Using Storybook and React, I prototyped queued offline uploads, dynamic floor plans, and validation check states. This allowed our engineering team to inspect edge-case behaviours—such as mid-sync dropouts—before final build phases.

Deep Dive: Key Design Decisions

Task-Driven Home: Centering the Daily Site Schedule

Spatially orienting the assessor. Rather than opening to a generic search or a list of active building structures, the homepage answers a critical question: what needs to be inspected next? It maps the daily schedule, surfaces site risk profiles, and provides spatial orientation via interactive floorplans.

Three mobile screens showing inspection home, schedule, and task map — Home, schedule, and task map were organised around today's inspection work.

Minimising flow friction. To answer clause questionnaires without losing their physical place on site, assessors use inline expansion cards. The sidebar copilot pre-arms assessors with common deviations and site history notes without dictating the final compliance decision.

Checklist list and checklist item detail with answer states, remarks, attachments, and defect links — Checklist detail kept answer states, remarks, attachments, and related defects close to the clause.

Matching the physical context. While a compact smartphone layout is ideal for handheld on-site documentation, large floorplan reviews require spatial context. We developed an iPad-specific dashboard that keeps architectural floorplans and checklists visible side-by-side.

iPad assessment form with floor plan, selected room, location type, and assessment category checklist — The iPad concept supported room-based assessment when spatial context mattered.

Review-First AI: Supporting Decision Makers Without Gating Authority

Anchoring AI output. The assessor takes a defect photo, dictates a quick voice comment, and reviews the AI’s suggested classification. The voice step provides a moment of cognitive pause, ensuring the assessor remains the ultimate decision-maker before official log entry.

AI defect processing, defect edit form, and suggested checklist review screens — AI suggested the record and checklist links, then waited for assessor confirmation.

Calm Offline UI: Designing Offline Mode as the Normal State

Preserving status transparency. Network failures are treated as standard operational behaviour. A persistent, calm sync indicator informs the assessor exactly which records are successfully synchronised, queued in local storage, or failed, preventing redundant field reviews.

Offline mode with persistent status indicators and sync queue — Offline state was persistent and calm. Every item showed whether it was queued, synced, or failed.

Dynamic Copilot: Scoping LLM Advice to Bounded Guidance

Preventing open-ended errors. We avoided open chat windows. Instead, the copilot uses bounded prompt chips that address specific query vectors: common defects on similar structures, compliance examples, or emergency escalation procedures.

Contextual chatbot with recommended inspection questions and common defect guidance — Recommended questions reduced blank-prompt friction and anchored AI responses to the current inspection.

Structured Signoff: Unifying Evidence for Handoff

Verifying before submission. The final submission summary organises structured findings, skipped clauses, and photo evidence. The signoff button activates only when all mandatory safety segments are verified, enabling assessors to sign and submit reports on-site.

Inspection checklist and report review screens before report submission — Report review brought findings, remarks, attachments, and defect evidence together before submission.

Impact & Outcomes

Engineering readiness achieved. The designs were compiled and successfully handed off to engineering after a twelve-month cycle. Usability test rounds validated major speed gains across core workflows.

72%

faster defect documentation

Down from an average of 12 minutes per defect to under 3.5 minutes in testing.

3.5 hrs -> 28 min

report compilation time

End-of-inspection reports moved from office work to on-site submission.

94%

task completion in testing

Across 12 assessors and 3 test rounds covering core inspection flows.

AI features designed

Defect detection, copilot brief, clause suggestions, and contextual guidance.

72% faster

Defect documentation became quicker in usability testing.

Average documentation time dropped from about 12 minutes per defect to under 3.5 minutes.

3.5 hrs -> 28 min

Report compilation moved closer to the field workflow.

Reports shifted from office work to on-site review and submission in testing.

94%

Final-round task completion across core inspection flows.

The testing sample covered 12 assessors across 3 rounds.

Review-first AI

AI features supported assessors without taking authority.

Defect detection, clause suggestions, copilot guidance, and draft language stayed editable and overridable.

Reflection

Earning trust incrementally. In high-precision government workflows, AI cannot act like a magical assistant. It earns credibility by supporting human decision-making in small, predictable moments, staying fully transparent when uncertain, and making overrides simple.

Designing for the worst state. Field tools succeed or fail on their resilience. If a network drop creates visual confusion or data-loss anxiety, assessors will revert to paper notebooks. Treating the offline disconnected state as a normal condition is what made the app viable in active construction zones.

BuildProof Smart Inspection Mobile App

Inspection work moved from scattered field notes to one reviewable mobile workflow.

A quick summary