Live demo · vision AI + public market data + analog rate-card

From a customer's photo to a confirmed, paid job — in under 90 seconds.

Lloyd Park — #1 fan of 1-800-GOT-JUNK?

Drop a job photo. A vision model identifies the items, estimates the volume, flags risk. Two pricing inputs blend into a single quote: public market data (1-800-GOT-JUNK, College HUNKS, Junk King, HomeAdvisor surveys) and a rate-card analog from 445 anonymized broker tickets. Once we train on Lloyd Park's own historical jobs, the custom model replaces both.

445
Bulk Tickets Analyzed
205
With Final Pricing
3
Pricing Tiers Discovered
~90s
Photo → Paid
End-to-end system architecture: Mira voice intake → lead capture → photo upload → Gemini vision → intake report → dispatch → payment
End-to-end · seven stations from inbound call to paid job
1
Call & Qualify

An always-on receiving secretary. Voice or text, customer's choice.

Inbound calls and texts hit a trained AI agent — Mira — that qualifies the lead, captures contact info, and texts a one-tap photo upload link. 24/7, zero hold time. The qualified lead flows straight into the vision pipeline below.

Mira AI receiving secretary: phone, SMS, and chat inputs route to lead capture, SMS photo request, and vision pipeline handoff
Mira · three input channels · three downstream handoffs
M
Mira
Lloyd Park receiving secretary · ElevenLabs voice + Gemini brain
(yes, we're #1 fans of 1-800-GOT-JUNK?)
Hours24/7
ChannelsVoice · SMS · Web chat
Voicebrowser default
Braingemini-2.5-flash
Trained on445 broker tickets
Handoff toPhoto intake → quote
Production system prompt

        
Tool schema (ElevenLabs / Twilio)

        
Live · Mira
Speed Model Voice
2
Photo Intake

Customer drops a photo. The model sees what's there.

Drag in any image of a junk-removal job — a curbside pile, a garage, an office cleanout. The vision model identifies items, estimates volume, flags hazards.

Photo intake flow: smartphone drop zone → vision AI lens → intake report card with detected items, volume, and recommended quote
Photo → vision AI → structured intake report
Try a sample job photo click any tile to load
…or drop your own photo
click to browse · paste with Ctrl+V · up to 5 images
Reading photo…
3
Dispatch

Schedule check across the crew. Auto-assign to the best fit.

The system scans every registered crew's calendar — current location, route load, certifications, vehicle capacity — and picks the right team for this job. Tap below to run the dispatch logic.

Dispatch board: grid of crew cards with availability bars, one card highlighted as auto-assigned, with distance and capacity overlays
Auto-assign · capacity × distance × availability
Today · Crew availability
Idle
[ready] Waiting for job assignment…
4
Notify

SMS, then voice confirmation. The customer doesn't lift a finger.

Once a crew is locked in, the system texts the customer their crew, ETA window, and a one-tap confirm link. If they don't reply within 2 minutes, an AI voice agent calls them.

Customer notification: SMS confirmation, 2-minute reply window, AI voice agent escalation, confirmation check
SMS first · voice agent escalates after 2-min silence
9:41●●● 5G
LP
Lloyd Park Dispatch
+1 (555) 010-4404
Voice agent · Standby
Awaits SMS timeout
Transcript will appear here…
5
Pay

Payment captured on-site. Receipt and invoice synced.

Crew completes the job, taps "done" in the field app. Customer's stored card runs automatically; receipt sends via SMS + email; the line item drops into accounting.

Job Total
$375.00
Bulk pickup · 1 truck · ~7 cu yd
🔒 PCI compliant Stripe · Square · Adyen ready
Receipt
Job IDLP-2026-00187
ServiceBulk pickup · Loose Item
Crew
Completed
Subtotal$375.00
Total charged$375.00
6
The Model

Today: public market + rate-card analog. Tomorrow: your custom pricing model.

The quote shown right now blends two inputs: 2026 broker market rates (incl. dump fees, labor, fuel) and an analog rate-card floor from 205 priced broker tickets. Once we ingest Lloyd Park's historical jobs — quote, win/loss, final billed, margin — we train a custom pricing model on your ground truth. It replaces both inputs and tunes to your conversion patterns, your geographic mix, and your target margin per job.

Training pipeline: 445 anonymized broker tickets feed feature extraction, rate-card classifier, and public market blend, producing an intake report with a recommended quote
Training pipeline · ground truth → features → blended quote
Public data sources
5
GOT-JUNK · College HUNKS · Junk King · LoadUp · HomeAdvisor
Analog tickets
445
Anonymized broker bulk-pickup work orders
With pricing
205
~46% of bulk tickets carry NTE
Custom GPT model
Anticipated · trained on Lloyd Park history

Public market reference rates

Quoted ranges from public junk-removal pricing pages, normalized to one 15-cu-yd truck. Used as the demo's first pricing input until your historical data replaces it.

Analog rate-card distribution

From 205 anonymized broker tickets where Not-To-Exceed was captured. Three deterministic tiers.

Tickets analyzed
445
Bulk-pickup work orders, 2026 to date
With final pricing
205
~46% of bulk tickets carry NTE
Pricing tiers
3
$205 · $225 · $375
Classifier accuracy
100%
Reason→tier is deterministic in this set

Pricing distribution by tier

From 205 tickets where Not-To-Exceed was captured.

By reason

What the customer said the job was for — and what it priced at.

Sample tickets

All identifiers anonymized. Real descriptions, real prices.

7
Rollout

What we ship for Lloyd Park.

First 30 days replicates this demo end-to-end on Lloyd Park's own data and accounts. 60-90 days adds the ML lift and the broker network expansion.

Week 1
Ingest Lloyd Park's historical job + invoice data
Same parser pattern we used on the source dataset — descriptions, items, location, final price, win/loss. Build the rate-card classifier from your real history.
Week 2
Vision intake live on your customer portal
Drop-photo widget → AI item detection → instant quote. Lloyd Park-branded, your domain.
Week 3
Crew scheduler + dispatch logic
Connect to your existing scheduling tool (or ship our calendar). Auto-assign by zone, capacity, certifications.
Week 4
SMS + voice confirmation flow live
Twilio for SMS and voice. AI confirmation agent for no-reply timeouts. Customer text-back capture.
Week 5–6
Payment + receipt + accounting sync
Stripe or Square based on your stack. Receipt SMS+email. Line items into QuickBooks / NetSuite.
Week 8–12
Margin model + provider scorecards
Train the second-stage model: predict margin per job, surface low-margin patterns, score sub-haulers on completion and customer rating.