LanePal Sales Tutorial

Left: LanePal Branding

• LanePal title and "Agent Intelligence Dashboard" subtitle
• Just branding. Nothing interactive here.

Center: Agent Toggle Buttons

• Four buttons, each representing a phone endpoint that can receive incoming calls:
• Andy Cell PSTN agent — calls ring Andy's cell phone
• Alt Phone PSTN agent — calls ring the alternate phone number
• OpenAI English OpenAI voice agent — AI answers in English
• OpenAI Spanish OpenAI voice agent — AI answers in Spanish
• Blue highlight = currently active. Click a different button to change which phone/agent receives the next call.
• This change syncs across all browsers viewing the dashboard via WebSocket. Everyone sees the same active agent.
• The active agent is stored in DynamoDB, so it persists across refreshes and backend restarts.

Right: Connection Status Badge

• Connected = WebSocket is live, dashboard is receiving real-time data
• Disconnected = WebSocket dropped. Dashboard will auto-reconnect. If it stays red, the backend may be down.

Call Metrics

• Active Calls — number of calls happening right now (updates live via WebSocket)
• Total Calls — cumulative calls since the backend was last restarted (recovered from DynamoDB on startup)
• Avg Priority — average priority score from AI classification across all calls (1-5 scale, where 5 is highest urgency)

Microphone Button

• A legacy feature from Mode A (browser-based transcription). Clicking it starts capturing audio from the browser's microphone and streaming it to the backend for transcription.
• Not needed for normal demos. The production system (Mode B) captures audio from the phone call via Chime Meeting + KVS. The mic button is a fallback.
• If someone asks about it: "That's our browser-based capture mode for testing. In production, we capture directly from the phone call — zero setup for the agent."

Model Badge: Nova Micro + Aurora Postgres

• Shows which AI model and database power the suggestions.
• Nova Micro is Amazon's fast, cheap inference model on Bedrock. Aurora Postgres is the load database.
• Two sequential AI calls: (1) extract intent + generate SQL, (2) summarize results into plain English. Total: ~1.5 seconds.

Proactive AI Area

• Default state: "Listening for dispatch requests on active calls..." with a search icon
• When the AI triggers, this area populates with:
• Tag pills — colored badges showing extracted parameters (e.g., From: Seattle To: Portland Reefer)
• Summary text — natural language description like "Found 7 loads from Seattle to Portland, rates ranging from $2.10 to $3.45 per mile"
• Results table — columns: Carrier Name, Rate/Mile, Total Rate, Distance, Weight, Commodity, Equipment Type, Broker Credit Score
• Suggestions update as the conversation evolves. If the caller changes from "Seattle" to "Tacoma," the AI re-queries automatically.
• Triggers: the AI analyzes every 6 transcript segments, or immediately when it hears high-intent keywords (city names, "load," "truck," "shipment," equipment types).

Manual Query Box ("Ask AI")

• Collapsed by default under a "Manual Query" toggle. Click to expand.
• Type any natural language question and press Enter or click "Ask."
• The AI converts it to SQL, queries the database, and returns results with the generated SQL visible.
• Good demo examples:
  • "Reefer loads under $3/mile from Portland"
  • "Which brokers pay within 7 days?"
  • "Flatbed loads from Boise to Spokane after April 20"
  • "What commodities are moving from Yakima this week?"
• This is the fallback if proactive AI does not trigger, or for showing a specific query on demand.

Call Cards

• Each ongoing call appears as a card showing: caller phone number, call status, and start time.
• Status badges: Ringing Transcribing Ended
• Click a call card to select it — the transcript and classification panels will show data for that call.
• When no calls are active, shows "No active calls" placeholder.

Transcript Stream

• Scrolling text with speaker labels: Agent (green) and Customer (purple)
• Updates in real time as people speak. Latency: 1-2 seconds from speech to text on screen.
• Auto-scrolls to the bottom so the latest speech is always visible.

Source Tabs: All / OpenAI / AWS Transcribe

• Three tab buttons at the top of the transcript panel.
• All — shows transcripts from any source (default)
• OpenAI — shows only transcripts from the OpenAI agent (when using an OpenAI voice agent)
• AWS Transcribe — shows only transcripts from Amazon Transcribe Streaming (Mode B phone audio)
• In a normal PSTN demo, all transcripts come from AWS Transcribe. The tabs matter when comparing transcription sources.

Call Classification

• After 4 transcript segments, GPT-4o-mini classifies the call. Fields shown:
• Category badge — Dispatch Inquiry Booking Complaint
• Priority score — 1 to 5 (5 = highest urgency). Drives the "Avg Priority" stat at the top.
• Summary — one-line description of the call's purpose
• Key entities — extracted names, locations, freight details
• Suggested actions — what the agent should do next
• Classification updates as more of the conversation is heard.

System Events

• Scrolling log of system events: WebSocket connections, call lifecycle events, AI triggers, errors.
• Copy Log button — copies the entire log to clipboard. Useful for debugging or sharing with the team.
• During a demo, this panel is usually ignored. It is there for transparency and troubleshooting.

Contact List

• Pre-loaded contacts: Andy (+14252307002) and Hawk (+16306398711) are loaded by default for testing.
• Each contact card shows: name, company, phone number, status badge, and notes.
• Status badges cycle through the call workflow: Pending Calling Connected then disposition-specific colors after the call ends.
• Hover over a contact and click the X to remove them from the list.
• The list persists in browser localStorage — survives page refreshes.

CSV Upload Area

• Drag and drop a CSV file or click to upload. Bulk-loads contacts from a spreadsheet.
• Expected columns: Name, Company, Phone, Notes.

Add Contact Form

• At the bottom of the left panel. Fields: Name, Company, Phone (e.g. +15551234567), Notes.
• Click "+ Add to List" to append to the dial list.
• Tip for demos: add the investor's phone number so they can experience receiving a call from the system.

Call Info Card

• When idle: "No active call. Start a session to begin dialing." with a phone icon.
• During a call: shows the contact's name, company, phone number, and any notes you entered.
• Call timer — counts up from 0:00 the moment the call connects.
• Status indicator text: IDLE before session starts, DIALING while ringing, CONNECTED when answered, CALL ENDED when finished.

Live Transcript

• Same real-time transcript as the inbound dashboard, but with Agent and Contact speaker labels.
• Appears once the call is connected and both parties are speaking.

AI Load Suggestions

• Same proactive AI engine as the inbound dashboard. As the conversation mentions cities, equipment types, or dates, matching loads appear automatically.
• Default state: "Waiting for active call to analyze..."
• When triggered: tag pills, summary, and results table — identical format to inbound.

Start Session / End Session

• Big green button with a pulsing animation when ready.
• Click Start Session to begin dialing the first contact on the list.
• Button changes to End Session (red) while a session is running.
• Ending the session hangs up the current call and stops auto-dialing.

Skip to Next Contact

• Skips the current contact and moves to the next one in the list.
• Useful when someone does not answer and you do not want to wait.
• Disabled when no session is active.

Progress Bar

• Shows "X of Y contacts" and a percentage completion bar.
• Only visible during an active session.
• Fills up as you move through the dial list.

Disposition Buttons

• Appear after each call ends. Four options:
• Interested — contact wants to proceed
• Callback — call back later
• Not Interested — no fit
• No Answer — did not pick up
• Setting a disposition marks the contact as complete and (if auto-dial is on) triggers the next call.
• Buttons are disabled until a call ends.

Auto-Dial Controls

• Toggle switch — turns auto-advance on/off. When ON, the next contact is dialed automatically after you set a disposition.
• Delay slider — configurable from 1 to 10 seconds. Sets how long to wait before dialing the next contact. Default is 3 seconds.
• The delay gives the agent time to take notes before the next call starts.

Session Stats

• Four metrics updated in real time during the session:
• Calls Made — total dials attempted
• Connected — how many calls were answered
• Avg Duration — average call length (mm:ss)
• Interested — contacts marked as "Interested"

Same as Inbound

• System events, WebSocket messages, errors. Copy Log button for debugging.

Left Side: Live Architecture Diagram (SVG)

• 16 components rendered as interactive nodes, connected by edges showing data flow:

Right Side: Event Ticker

• Raw events stream in with timestamps and highlighted keywords.
• Shows the actual data flowing through: transcript snippets, AI triggers, classification results.
• During a live call, you can see each spoken phrase flow from Transcribe through the Backend to the Dashboard.

How to Use It

• Click "Simulate Load" to start the simulation.
• Speed control buttons appear: 1x 5x 20x 50x
• 1x — realistic pace. A few calls flowing through, components light up one at a time. Good for explaining each step.
• 5x — busier. Multiple components lit simultaneously. "This is a moderately busy dispatch office."
• 20x — architecture is constantly lit, events flowing rapidly. "This is a busy Monday morning."
• 50x — everything is blazing, events/sec counter climbs, cost estimate appears. "This is the same infrastructure handling 50 concurrent calls. No hardware changes, no trunk upgrades. AWS scales it automatically."

Stats Displayed During Simulation

• Events count — total events processed
• Events/sec — current throughput rate
• Concurrent calls — how many simultaneous calls are being simulated
• Estimated monthly cost — rough dollar figure based on the current load level

Four Fictional Companies (Tenant Cards)

• Each company has its own phone number, its own calls, its own event stream.
• But they all share one infrastructure — shown by the shared architecture diagram in the center of the page.

Shared Architecture Diagram

• Same SVG diagram as the DEBUG panel, but placed prominently in the center.
• Components light up with events from ALL four tenants — demonstrating shared infrastructure.
• A global event ticker at the bottom shows a live stream of events across all companies.

Header Stats and Speed Controls

• Concurrent Calls — across all 4 tenants
• Total Events — cumulative events processed
• Events/sec — current throughput
• Speed buttons: 1x (realistic), 5x (busy day), 10x (at scale)

• "One phone number per client — not per agent, not per call." Each company gets a single inbound number. Concurrent callers share it.
• "Zero idle cost — no calls means no charges." Per-minute billing. When nobody is calling, the bill is zero.
• "Adding a new client: 3 API calls, 30 seconds, ~$1/month for the phone number." No hardware, no contracts, no provisioning delays.
• "No hardware, no trunks, no PBX, no contracts." Every call is a temporary software object created by an API call.
• "Same infrastructure that handles 1 call handles 10,000." AWS auto-scales. No capacity planning needed.

Current Database (Simulated)

Geographic Coverage

• Pacific Northwest: Washington, Oregon, Idaho, Montana
• Key cities: Seattle, Portland, Spokane, Boise, Yakima, Tacoma, Bellingham, Eugene, Bend, Medford, Missoula, Billings, Olympia, Kennewick, Wenatchee, Idaho Falls, Salem
• Date range: March 23 through April 30, 2026

Fields the AI Can Query

• Origin and destination cities
• Rate per mile and total rate
• Weight and capacity (FTL, LTL, Partial)
• Equipment type (dry van, reefer, flatbed, stepdeck, tanker)
• Commodity description
• Refrigerated status and temperature requirements
• Broker credit score and days to pay
• Trailer length
• Hazmat certification
• Pickup and delivery dates

What DAT Integration Adds

• Hundreds of thousands of live loads (nationwide, not just PNW)
• Real-time market rate context from DAT RateView ("This load is above/below market rate")
• Load freshness ("Posted 2 hours ago" vs. "Sitting for 3 days — broker may negotiate")
• Deadhead calculation from the truck's current location
• Verified broker credit scores and days-to-pay
• 60+ equipment type codes (vs. our current 5 primary types)