AR Site Inspector

Overlay the plan, photograph the defect, report the snag

AR Site Inspector is a field tool for construction inspectors. An inspector signs in, sets up a project and uploads its building plan (an image exported from a DWG or PDF), then opens the AR inspection view to overlay the plan on the real site. On site they photograph a condition or defect — the photo is captured with the device camera and tagged with the device's geolocation. A vision AI assesses the photo for construction compliance: it describes the observation, states whether it looks compliant, classifies the deviation type, assigns a severity (critical, major or minor), cites a relevant IS-code where one applies, and suggests a corrective recommendation. The inspector reviews and edits that into a snag, saves it, and can ask the AI to draft a non-conformance report (NCR) for it. The snag list tracks each snag from open to closed, and a geotagged PDF snag report pulls it all together — project details, severity counts, an AI executive summary and a snag-by-snag section with the geotagged photos embedded. The tenant is the firm: every inspector shares that firm's projects, plans, snags, photos and reports, and each record is scoped to the firm.

Please read this before relying on any result

Built for the people who inspect a build

It suits any construction firm that wants to standardise a site inspection and produce a clear, photo-backed snag report. AR Site Inspector is a decision-support and documentation aid: the AI reasons over the photo and notes the inspector provides — it does not replace a qualified engineer's judgement, and every result must be independently checked before it is relied on.

Per-project subscription

AR Site Inspector is billed as a flat fee per construction project — one price per project covers project setup and plan upload, the AR plan overlay, AI photo compliance assessment, the snag list with AI-drafted NCRs, and geotagged PDF snag reports for the whole team on that project. Payment is via Razorpay from the Subscription area, and a successful charge keeps the firm plan active for one project cycle from the payment date.

The page map

From a building plan to a geotagged PDF snag report

Four focused AI passes, all routed through ASI_ClaudeHelper

All AI calls are routed through ASI_ClaudeHelper and never throw — on any error or malformed reply, the calling page shows a clear message instead of a raw exception. The provider is configurable (Groq by default, or the Claude API), the photo assessment uses a vision model (meta-llama/llama-4-scout-17b-16e-instruct), and every prompt returns strict JSON so the pages can render structured, defensively-parsed results. None of these outputs is a certified inspection; each is an aid for a qualified engineer's own assessment.

AssessCompliancePhoto — vision photo compliance

You are assisting a qualified site engineer with a first-pass
construction compliance check of a single site photo. This is
decision support, NOT a certified inspection. Assess ONLY what is
visible in the photo plus the context provided.
Project type: "{{project_type}}". Location context: {{location}}.
Inspector note: {{inspector_note}}. [image attached]
Return STRICT JSON only:
{ "observation": "what is visible in plain language",
  "complianceStatus": "compliant | non-compliant | needs review",
  "deviationType": "e.g. Honeycombing | Cover deficiency | Crack |
     Alignment | Housekeeping | Safety | Finishing",
  "severity": "critical | major | minor",
  "isCodeRef": "a relevant IS code if one clearly applies, else ''",
  "recommendation": "a corrective action to consider" }
Base it only on the photo and context. If the photo is unclear, say
so and use 'needs review'. A qualified engineer must verify every
point. No emojis, no hype.

ClassifyDeviation — classify a described defect

Classify this described construction observation for a
"{{project_type}}" project. This is decision support, NOT a certified
inspection. Description: {{description}}
Return STRICT JSON: { "deviationType": "short category",
  "severity": "critical | major | minor",
  "isCodeRef": "a relevant IS code if one clearly applies, else ''",
  "recommendation": "a corrective action to consider" }
Base it only on the description. A qualified engineer must verify it.
No emojis, no hype.

GenerateNCR — draft a non-conformance report

Draft a concise non-conformance report (NCR) for a construction snag,
as decision support for a qualified engineer to review and issue.
Snag title: {{snag_title}}. Description: {{snag_description}}.
Deviation type: {{deviation_type}}. IS code ref: {{is_code_ref}}.
Return STRICT JSON: { "ncrTitle": "short NCR title",
  "nonConformance": "what does not conform and why",
  "requirement": "the requirement or standard it should meet",
  "correctiveAction": "the corrective action proposed",
  "note": "a reminder that an engineer must confirm and sign off" }
Base it only on the details given. No emojis, no hype.

SummariseSnagReport — an executive summary

Summarise this project's snag list for a site snag report, as
decision support for a qualified engineer. Project: {{project_name}}.
Snags (JSON): {{snags_json}}
Return STRICT JSON: { "summary": "a short plain-language summary of
  the key findings and the highest-severity items",
  "priority": [ "the most important items to address first" ] }
Base it only on the snags provided; do not invent counts. Remind that
an engineer must verify the findings. No emojis, no hype.

An orbit-controlled 3D plan view, with a WebXR plan overlay where the device supports it

Plans are exported images, not parsed DWG files

Critical, major and minor — used consistently everywhere

Every snag carries a single, fixed-colour severity. The same three colours are used on the snag cards, the snag list, the report table, the donut chart and the PDF, so the meaning never shifts.

The stack

Shared helpers referenced across the pages: ASI_DBHelper (GetTable / ExecuteSP / Scalar / P), ASI_ClaudeHelper (AssessCompliancePhoto, ClassifyDeviation, GenerateNCR, SummariseSnagReport), ASI_WhatsApp, and the shared AIS RBAC/auth kit for sign-in, since there is no separate ASI_Auth table — the inspector's identity comes from the shared kit and the firm is the OrgId.

A firm-scoped data model

Tables

Key stored procedures

Deploy via sqlcmd

Apply the product schema directly to the shared SQL Server instance, then verify object counts. The script is idempotent and prefix-scoped — it never touches other products' objects. The prefix is ASI_.

sqlcmd -S sql5083.site4now.net -U db_user -P ****** -d db_9b0fff_aiscompany ^
       -i C:\asp.net\CompanyProjects\AIAis\ArProjects\ARSiteInspector\ASI_Database.sql

-- verify
SELECT COUNT(*) FROM sys.tables     WHERE name LIKE 'ASI_%';     -- the product tables
SELECT COUNT(*) FROM sys.procedures WHERE name LIKE 'ASI_usp_%'; -- the proc layer

Serve the site over HTTPS at /ArProjects/ARSiteInspector/ — required for the camera and geolocation used by the AR overlay, the QR-anchor scanner and geotagged capture. Set the Groq/Claude and Razorpay credentials in the shared config. Sign in, set up a project and upload its plan image, open the AR inspection to overlay the plan on the real site (or review it in 3D), capture a defect photo and run the AI compliance assessment, save the snag and draft an NCR, then download the geotagged PDF snag report.

Demo & review accounts

How projects, photos and the AI results are handled

Where AR Site Inspector goes next

• Plan-pinned snags: pin each snag to an exact point on the plan overlay, so the snag list can be read as a heat-map on the drawing, not just a list.
• WebXR hit-test measurements: use the AR hit-test to offer approximate on-site distances as an aid, clearly flagged as indicative and not survey-grade.
• Round-over-round comparison: compare a snag list across inspection rounds to show what was fixed, what re-opened and what is still outstanding per stage.
• Project & stage NCR packs: curated deviation checklists and standard IS-code references per project type and construction stage, maintained by the firm.
• Engineer sign-off workflow: a review-and-approve step where a qualified engineer confirms or overrides each AI assessment and severity before a snag or NCR is finalised, reinforcing that the engineer makes the final call.
• Offline capture & sync: capture geotagged photos and snags with no signal on site, then sync the round when the device is back online.
• Contractor snag portal: a read-and-respond view for a contractor to mark a snag as addressed with a fix photo, before the inspector re-inspects and closes it.