See exactly what is
really happening on site
Tekshot Vision AI captures what is happening on site as photos and video, analyses them to extract evidence, then sends that evidence straight into Output. The value is not in being able to view a photo — it is in the photo becoming the basis for acceptance and payment.
- Captured on a basic phone
- Results always come with a confidence level
- The final verifier
Real evidence
- Photos & video with GPS coordinates and timestamps
- The AI suggests the work item and an estimated quantity
- Engineering confirms it before it becomes evidence
- Evidence feeds Output; it does not stop at a photo library
We do not claim the AI is always right
The people who evaluate the product inside a contractor are engineers. They trust a system that knows how to say it is unsure more than one that is always right. So every Vision result in ERPCons is shown confidence, and every case the machine is unsure about is routed to a human approver instead of being decided automatically.
- No demos on a hand-picked photo set
- We do not hide misrecognition cases
- AI results are never used as a basis for payment without human approval
From the site to output
This is the whole journey of a single site photo. Step 4 — verifier — is mandatory and cannot be switched off.
Stopping at step 3 leaves you with a smart photo library. The real value is in step 5: site evidence flowing into Quantity & quality acceptance, then continues on to Cost and Reconciliation.
The real interface (sample data)
| Photo | Work item suggested by the AI | Location | Estimated qty | Confidence | Status |
|---|---|---|---|---|---|
| Slab concrete pour | Level 5 — grid A/1-3 | ~ 42 m³ | 94% | Fast-track approval | |
| Beam rebar fixing | Level 5 — grid B/2 | ~ 3.1 tonnes | 91% | Fast-track approval | |
| Blockwork | Level 4 — grid C | ~ 68 m² | 72% | Needs a human approver | |
| Column formwork (2 options) | Level 5 — grid D/4 | — | 58% | Needs a human approver | |
| Indeterminate — backlit photo | Level 5 — grid E | — | — | The AI draws no conclusion |
Interface shown with sample data. The 65% / 26% / 9% ratios are illustrative — not yet a measured benchmark. See §Limitations below.
Confidence & technical limits
Under the evidence principle (ERPCONS-WEB-UX-001 §10, item 10), the page must not publish AI capabilities without a benchmark. The content below is a frame to fill in, not measured figures. Do not publish this page before replacing them with real data.
Three levels of results
Conditions where the AI performs poorly
Published openly so users know up front, instead of discovering it after rollout.
- Backlit shots, or night shots without enough light
- The angle is too far away, or most of the work item is obstructed
- The work item is mid-transition and not clearly defined yet
- The material or construction method is not in the training set
- [Awaiting Product to add the conditions actually recorded]
What Vision AI no do
- Output is never recorded automatically without an approver
- It does not replace acceptance against technical standards
- It does not judge internal structural quality from a surface photo
- It does not replace human supervision on site
Proven use cases
Awaiting Product sign-off on the listWe list only cases that ran on real projects with measurable results. The four cards below are a template — each card needs its scope, input data and measured result filled in.
Replace manual progress reports with time- and location-stamped photos.
Photos attach automatically to the right work item and position on the drawing.
Detects missing PPE and hazardous areas without barriers.
Compare photos of the same spot across periods to see the progress gap.
Runs on the devices actually used on site
The main users are site managers and field engineers on mid-range Android phones with unreliable mobile coverage. That is a design constraint, not a preference.
- Capture and upload works on a mid-range Android phone
- Works offline, syncs automatically once there is signal
- Large touch targets that work with gloves on
- Images are compressed on the device to save bandwidth
Security — Transparency — Compliance — Ready to scale
Want to try it on your own site photos?
A Vision pilot runs on one project's real data with a clear scope and success criteria — a committed trial, not a demo.
- An expert works alongside you
- Scope and criteria agreed up front
- Your data belongs to you