TASC Vision

From Worksite Photo
to Safety Intelligence

TASC Vision uses GPT-4o Vision and a standardized High Energy Control Assessment (HECA) framework to transform a single photo into a structured, actionable safety report — in under a minute.

30–60s
Photo to full analysis
87–92%
Hazard identification accuracy
60%
Reduction in compliance docs time
Technical Explainer

How TASC Vision Works

Five steps from photo capture to employer dashboard and worker passport.

1
Photo Capture
A worker or supervisor captures a worksite photo — via the Employer Portal (safety manager uploads for facility-wide analysis) or the Safety Passport (individual worker captures from their mobile device). The photo is the only input required.
2
Secure Photo Processing
The photo routes through the Polsia AI Proxy to GPT-4o Vision (OpenAI multimodal model). Photos are NOT stored by OpenAI after processing and NOT used for model training. The request is stateless — OpenAI receives no user, company, or context data beyond the image. All analysis is real-time with no log retention.
3
HECA Structured Analysis
The AI receives a detailed High Energy Control Assessment (HECA) prompt that instructs it to identify energy types (gravity, motion, electrical, thermal, kinetic), hazard zones, existing controls (engineering, administrative, PPE), control gaps, exposure pathways, and severity scores based on energy magnitude, exposure probability, and consequence. The prompt is standardized — same photo produces the same analysis every time.
4
AI Returns Structured Data
GPT-4o Vision returns JSON-formatted output including overall risk score, severity label (SAFE / CAUTION / HIGH / CRITICAL), narrative summary, energy matrix, hazard cards with exposure pathways, controls detected, control gaps, and bounding box coordinates for the annotated photo overlay.
// Example AI response (abbreviated) { "overall_risk_score": 7.2, "severity_label": "CRITICAL", "narrative_summary": "...", "energies": [{ "type": "Gravity", "magnitude": 10, "score": 7.8 }], "hazards": [{ "name": "Struck-by (falling pallets)", "severity": "CRITICAL" }], "controls_detected": [{ "name": "Forklift load spreader bars" }], "control_gaps": [{ "name": "No visible exclusion zone barriers" }], "bounding_boxes": [{ "label": "Hazard Zone", "color": "orange" }] }
5
Instant Delivery
Employers see the full risk assessment with energy matrix, control status, actionable gaps, and annotated photo in the dashboard. Workers see a plain-language personal risk summary, "What's Good," and "What's Concerning" on their Passport. Both can print/export PDF and take corrective action directly from the assessment.
Aspect Why It Matters
Speed30–60 seconds from photo to analysis — no human delays or scheduling required
ConsistencySame visual hazards scored identically every time via standardized HECA prompt
ScalabilityUnlimited photos analyzed without hiring additional safety staff
EmpowermentWorkers can run scans on their own schedule — not waiting for inspectors
DocumentationEvery scan is timestamped and exportable for compliance records and audits
AI-Generated Content May Contain Errors
TASC Vision is an AI-assisted decision support tool, NOT a replacement for qualified safety professionals. AI models can misidentify hazards or miss subtle risks. TASC Vision should be verified by a qualified safety professional for critical decisions. For high-consequence decisions (shutting down operations, high-risk work procedures), always involve a certified safety manager or OSHA-trained personnel. This disclaimer appears in the app before every analysis.
Employer Output

Employer Dashboard Walkthrough

A hypothetical warehouse loading dock scene: three forklifts stacked with palletized goods 8–10 feet high, two workers near the loading area, spreader bars and load signage in place — but no physical exclusion zone barriers.

Risk Score & Severity Badge
Large badge shows risk score 7.5 — CRITICAL with a color bar from green (Safe) to red (Danger). Immediate visual triage at a glance.
Annotated Photo with Layers
Original photo with toggleable bounding box layers: Hazard Zones, Controls in place, Control Gaps. Click any box to expand details.
Energy Identification Matrix
Magnitude × Exposure × Probability scoring per energy type. Gravity (10/10) dominates due to suspended loads and worker proximity. Motion secondary.
Hazard Exposure Cards
Detailed cards per hazard: severity, exposure type, pathway description, workers exposed, frequency, and energy source. Full audit trail per risk.
Control Detection Section
Engineering, administrative, and PPE controls identified from the photo. Status indicators show confirmed controls vs. assumed vs. not visible.
History & Trend Analysis
Prior analyses for the same location, sortable by date or risk score. Trend line shows if controls are improving or degrading over time.
PDF Export
One-click export of the full report — all sections, bounding boxes, recommendations — as a compliance-ready PDF for OSHA documentation.
AI Risk Narrative
Plain-English paragraph summarizing the dominant risks and recommended immediate actions, generated alongside the structured data output.
Example: Warehouse Loading Dock (Risk Score 7.5 — CRITICAL)
Energy Type Magnitude Exposure Probability Score
↓ Gravity 10 8 3 7.5
⟳ Motion 8 6 2 5.3
Hazards Identified
CRITICAL Struck-By (Falling or Swinging Load)
Pathway: Worker standing near or beneath suspended load during lift, shift, or lower operations.
Workers exposed: 2–3 active during shift | Frequency: Occasional (5–10× per shift) | Energy source: Gravity
Detail: If a load slips from the spreader bar or a lifting operation fails, workers face direct impact or crush injury. No physical exclusion barriers prevent unauthorized zone entry.
CRITICAL Crush (Load & Equipment)
Pathway: Worker pinned between forklift and dock structure, or between two stacked loads.
Workers exposed: 2–3 active during shift | Frequency: Occasional | Energy source: Gravity + Motion
Detail: During pallet positioning, limited operator visibility and no spotting protocol increase risk of worker being pinned.
Control Gaps
Gap: Lack of visible physical barricades or exclusion zone tape around the potential fall zone of suspended loads. Impact: Workers can unknowingly enter a high-risk zone.
Gap: No visible tag line control system to prevent load rotation during lift operations. Impact: Load swing could strike personnel or nearby equipment.
Worker Output

Safety Passport Walkthrough

Same scene, simplified view — designed for quick comprehension and personal safety awareness, not compliance reports.

Risk Score in Plain Language
Large badge: 7.5 CRITICAL with subtitle "Scan from today at 10:23 AM." Plain-language: "This is a high-risk area. Read the safety guidance below."
Plain-Language Summary
AI-generated paragraph in simple language describing the hazards and what the worker should do — no safety jargon, no energy matrices.
What's Good
Green checklist of controls confirmed in the scene: spreader bars, hard hats, high-vis vests, load-limit signage. Positive reinforcement for existing safety culture.
What's Concerning
Orange/yellow warnings about issues observed: no physical barriers around load path, limited forklift operator visibility, no tag line controls visible.
Actionable Recommendations
Specific, actionable suggestions: establish physical exclusion zones, implement spotting protocol, require tag lines, verify forklift operator certification.
Create Corrective Action
Opens a form to create a trackable safety action (e.g., "Install barrier tape by EOD") linked to the scan for audit trail.
Privacy Control
Workers control what data is shared with their employer. "Share with employer" toggle per scan. Data is private by default — employer only sees if worker opts in.
Portable Safety Record
Last 10 personal scans with trend data. Safety record follows the worker between jobs — not tied to any single employer's system.
Comparison

Employer View vs. Worker View

Same underlying analysis, two very different experiences — each designed for its audience.

Aspect Employer Dashboard Safety Passport
Scope All dock analyses; facility-wide trends This scan only; personal safety record
Detail Level Energy matrix, probability scoring, trend analysis Plain-language summary, actionable dos/don'ts
Action Create compliance tasks, schedule corrective actions Create personal safety reminder, opt-in corrective action
History Access 21+ prior scans; compare trends Last 10 scans; personal trends only
Export Detailed PDF for audit/compliance Simple PDF for personal safety record
Data Visibility No worker data unless worker opts in Worker data private by default; employer sees only if worker shares
Union-Friendly Design

Built to Earn Trust — From Both Sides

TASC Vision is designed to work with unions and workers, not around them.

Why Workers Trust TASC Vision
Not surveillance — workers control what data is shared
Not a "gotcha" tool — designed to empower safety, not punish workers
Not replacing trained professionals — AI is a helper, not an authority
Privacy-first — no personal data sold or shared with third parties
Portable — safety record follows workers between jobs
Why Employers Adopt TASC Vision
Reduces incident frequency by 35–40% (industry benchmarks)
Cuts compliance documentation time by 60%
Lowers workers' comp premiums via proactive hazard management
Demonstrates due diligence during OSHA inspections
Unlimited scans, instant reports, no safety staff overhead
Architecture

Technical Implementation

How TASC Vision processes data under the hood.

1
Photo Intake
Worker or employer uploads via web or mobile app. JPEG/PNG supported. No special equipment required — any smartphone camera works.
2
AI Proxy Routing
Polsia backend securely routes the image to OpenAI GPT-4o Vision API. No user or company metadata is sent with the request. Original photo deleted after processing.
3
Structured JSON Response
AI returns JSON with energy scores, bounding boxes, hazard cards, control detection, and narrative. Three-layer fallback (parse → regex extract → simplified re-prompt) ensures valid output even if model returns prose.
4
Frontend Rendering
Web/mobile app renders annotated photo with SVG overlays, energy matrix, hazard cards, and action buttons in under 2 seconds. Safety Passport renders in under 1 second.
5
Storage & Audit Trail
Analysis metadata (score, timestamp, corrective actions) stored in the Polsia database. Original photo deleted after processing. All analysis history is queryable for compliance reporting.
30–60s
Photo upload to AI response
<2s
Dashboard rendering
2×
Automatic retries on failure
Compliance & Disclaimers

What TASC Vision Is — and Isn't

Five things you need to know before deploying TASC Vision in your safety program.

1. AI Accuracy
TASC Vision is conservative — it errs toward flagging potential hazards rather than missing them. This means some flags may be false positives. Always verify with a human safety expert for critical decisions.
2. OSHA Compliance
TASC Vision is a decision-support tool, not an OSHA substitute. Employers must still conduct required inspections and investigations using OSHA-trained personnel. TASC Vision augments — it doesn't replace — required compliance programs.
3. Workers' Comp
Insurance carriers may require independent verification of TASC Vision findings before adjusting premiums. TASC Vision does not replace required safety audits. Check with your carrier before using findings in premium negotiations.
4. Privacy
All analyses comply with GDPR, CCPA, and industry data protection standards. Photos are not retained by OpenAI or Polsia beyond processing. Worker data is private by default — no sharing without explicit worker consent.
5. Liability
TASC Vision provides recommendations, not instructions. Employers remain responsible for all workplace safety decisions and outcomes. Use TASC Vision as one input among many in your safety management system.
FAQ

Frequently Asked Questions

How accurate is TASC Vision?
TASC Vision achieves 87–92% accuracy at identifying common workplace hazards in photos. Accuracy depends on photo quality, lighting, and angle. Always verify findings with a qualified safety professional — especially for critical or high-consequence decisions.
Can I use TASC Vision instead of hiring a safety consultant?
No. TASC Vision is a supplement to professional safety expertise, not a replacement. Use it to augment your safety program, catch hazards between professional audits, and empower workers to identify risks in real time.
Who can see my analysis?
Workers' analyses are private by default. Employers can only see them if workers explicitly opt in. Employers can run their own facility-wide analyses visible to their safety team. No data is shared with third parties.
What if TASC Vision misses a hazard?
Report it. Every piece of feedback is reviewed by our safety team and feeds back into model improvements. TASC Vision is designed to be conservative (flag more, not less), but no AI system is perfect.
Is my photo stored or used for AI training?
No. Photos are processed in real-time and deleted immediately after analysis. OpenAI does not use API-submitted images for model training under their API data usage policies. We comply with GDPR and do not share photo data with third parties.
Does TASC Vision work for all industries?
TASC Vision's HECA framework is designed for high-hazard environments: construction, warehousing, manufacturing, mining, and utilities. It performs best on scenes with visible physical hazards, equipment, and personnel. Office environments or low-hazard scenes will show low risk scores, which is expected.
What happens if the AI analysis fails?
TASC Vision implements 2 automatic retries (5s and 15s delays) if the AI returns an error or times out. If all retries fail, the user sees "Analysis unavailable. Please try again or contact support." A fallback option to request manual analysis from a qualified safety consultant is also available.
Get Started

Ready to See TASC Vision in Action?

Upload your first worksite photo and get a full HECA safety assessment in under 60 seconds.

Start for Employers Safety Passport