FleetIQ

Freight dispatch still runs on phone calls and spreadsheets. With FleetIQ everything between load booked and 'invoice paid' now runs itself. Drivers tap once, Brokers get notified. 10X the Freights.

Comment

Inspiration

Trucking dispatch runs on fragmented emails, spreadsheets, and lenghty manual copy-paste. Our Co-Founder Luis Bermudez and his dispatch team process hundreds of freight documents daily: multi-page rate confirmations, broker credit checks, bills of lading (BOL), and proof of delivery (POD) receipts, manually typing dozens of data points into Google Sheets while trying to keep drivers moving.

A single missed address note or delayed rate confirmation can cost thousands in deadhead miles or lost revenue. We built FleetIQ to transform load intake and operations from a chaotic human bottleneck into an autonomous, AI-driven dispatch agent.

What it does

FleetIQ is an end-to-end, AI-powered Freight Dispatch and Operations Agent built for freight dispatchers, back-office accounting teams, and independent owner-operators.

  • Multimodal AI Load Intake: Dispatchers upload rate confirmations (PDFs, scans, images) or paste raw text. Powered by Google Gemini 3.5 Flash, the system uses spatial vision to extract multi-stop pickup/delivery schedules, rates, mileage, commodities, and billing instructions.
  • Autonomous 6-Step Workflow Orchestrator:
    1. Locates available rows in the master Google Sheet and generates sequential shipment IDs (HP-LB-00123).
    2. Ingests 21 structured fields validated against a live carrier/driver directory.
    3. Mirrors financial and invoice data to a 35-column Back-Office Financials Ledger.
    4. Automatically provisions, nests, and standardizes folders in Google Drive ({ShipmentID} - {BrokerLoadNumber}).
    5. Syncs driver assignments to the live Dispatcher Operations Board.
    6. Discovers broker email threads in Gmail and sends automated confirmation replies using pickup-aware smart templates.
  • Interactive Operations & Financials Dashboard: Real-time 5-stage Kanban board (Unassigned Pool, Dispatched, In Transit, Paperwork Pending, Completed), live fleet RPM ($/mi) tracking, detention/layover claim generators, and factoring status reconciliation (PENDING FACTOR → PAID).
  • Driver Self-Service Mobile Portal: Dedicated token-gated and email-authenticated mobile web portal where drivers view assigned route details, submit one-tap status updates, and take photos of signed POD/BOL paperwork (automatically compiled into PDFs and filed in Google Drive).
  • Security Audit & Guardrails: Built-in security engine auditing prompt injection risks, CORS configurations, outbound email blocklists, and sanitized PII logs.

How we built it

  • AI Engine: Google Gemini 3.5 Flash (with Gemini 3.5 Flash Lite fallback) via @google/generative-ai for multimodal document vision, spatial table extraction, document classification, payment remittance parsing (Square/OTR Factoring), and context-aware email drafting.
  • Frontend: React 19 + Vite + TailwindCSS 4 + Motion, with responsive mobile drawer navigation and touch-optimized operations boards.
  • Backend & Document Pipeline: Node.js 22 + Express hosted on Google Cloud Run (fleetiq-dispatch). Document parsing engine combining pdfjs-dist, pdf-to-img, sharp, pdf-lib, and tesseract.js for scanned document OCR.
  • Google Cloud & Workspace Services: Google Sheets API, Google Drive API, and Gmail API operating under a least-privilege service account with domain-wide delegation.
  • Auth & Database: Supabase PostgreSQL with Google OAuth (PKCE flow), granular Role-Based Access Control (admin, dispatcher, backoffice, driver, pending), and JWT validation on every API request.

Challenges we ran into

  • Multi-Stop Spatial Tables in PDF Rate Confirmations: Rate confirmations place pickup and delivery dates in arbitrary, multi-column layouts across several pages. Pure text OCR frequently jumbled stop sequences. We solved this by implementing Gemini Multimodal Vision (extractDataWithVision), passing high-resolution page renders directly to Gemini so it can understand spatial relationships and right-aligned timestamps natively.
  • Personal Gmail vs. Workspace Automation: Personal Google accounts lack Admin SDK and domain-wide delegation. We designed a hybrid service account architecture paired with a one-time OAuth consent bootstrap flow to ensure secure file-level sharing without granting over-privileged project IAM access.
  • Node.js 22 WebSocket Compatibility: The Supabase JS SDK requires native WebSocket support, which silently caused intermittent network failures on older Node runtimes. Diagnosing Cloud Run container logs led us to standardize our backend container runtime on Node.js 22.
  • Outbound Email Domain & PII Protection: Automating broker email replies risks leaking driver PII or emailing unauthorized addresses. We engineered a strict recipient sanitizer, domain allowlist filter, and server-side log scrubber to guarantee PII protection in terminal outputs.

Accomplishments that we're proud of

  • True Human-on-the-Loop Automation: Reduced rate confirmation processing time from 12–15 minutes down to under 15 seconds, with a 1-click preflight verification step.
  • Zero-Credential Production Security: Configured Application Default Credentials (ADC) on Google Cloud Run with Google Secret Manager injection — zero hardcoded JSON keys or secrets in source code.
  • Complete End-to-End Document Lifecycle: From initial PDF rate confirmation parsing to Google Sheet logging, Drive filing, driver dispatch, mobile POD upload, and broker reply emails — all unified in one pipeline.
  • Codebase Knowledge Architecture: Visualized the entire system architecture with Graphify (617 nodes, 1,150 edges, 36 modular communities) to ensure zero architectural drift during continuous development.

What we learned

  • Gemini's Multimodal Vision Outperforms Traditional OCR: Passing rendered document pages directly to Gemini 3.5 Flash resolved edge cases that broke regex and traditional OCR pipelines (e.g., watermarked PDFs, multi-column tables, concatenated RTS Pro MC numbers).
  • Attached Service Accounts & ADC Over API Keys: Attaching a dedicated service account directly to Cloud Run eliminated key rotation overhead and credential leak vulnerabilities.
  • Gated Onboarding with Pending Roles: Using Supabase's pending role state provides a bulletproof way to gate enterprise OAuth signups without building complex custom invitation systems.

What's next for FleetIQ

  • Full PostgreSQL Multi-Tenancy Cutover: Transitioning from Google Sheets to an isolated multi-tenant Supabase schema with Row-Level Security (RLS) and team management (per our 5-phase migration roadmap).
  • Direct Factoring API Integrations: Webhook-based integrations with TriumphPay and OTR Capital to automate payment reconciliations directly without intermediary PDFs.
  • Telematics & ELD Integrations (Samsara / Motive): Live GPS tracking and geofenced arrival/departure triggers to automate driver status updates.
  • AI Rate Negotiation Assistant: Leveraging Gemini to analyze historical lane rates and market averages to draft real-time counter-offers directly in broker email threads.
Share this project:

Updates