Country: United States · Website: www.dollartraq.com
The Challenge: Freight Brokerage Runs on Trust, and Trust Was Manual
US freight brokers lose money and time to three things: fraud and double brokering, manual carrier vetting, and back office work buried in email. Vetting a carrier means pulling FMCSA records, chasing insurance certificates, checking authority and insurance, then repeating it all at dispatch. Rate quoting, load tracking, and proof of delivery collection sit in inboxes. Existing freight tools rent their data from vendors and stop at "this carrier looks fine."
Dollar Traq wanted something far more ambitious: an AI operating system that sits on top of a broker's existing TMS and automates the entire trust layer, from first vet to final payment, while keeping a human in control of every consequential action.
The ProMonkey Solution: Five AI Agents on an Owned Data Foundation
ProMonkey Technologies engineered the full platform, from the data warehouse to the voice agents brokers talk to every day.
The data foundation
- A self hosted PostgreSQL warehouse holding 22.7 million FMCSA rows covering roughly 1 million carriers: authority history, crash records, inspections, out of service orders, and insurance filings. Queried directly, never rented through a third party API.
- Over 350,000 carrier insurance certificates on file, read by AI and verified with the certificate holder.
Trust Score v3, the vetting engine
- A 149 rule additive scoring catalog on FMCSA data, with exposure normalized crash and violation math (per power unit, per inspection).
- Hard stop gates aligned to CAVRA that brokers cannot disable, and data confidence caps so thin file carriers can never score Preferred.
- Every score shows its inputs, weights, and the government records behind them. No black box numbers.
The AI agent layer
- Sentinel: carrier trust and fraud screening. Chameleon carrier detection, domain age checks, lookalike alerts, and 24/7 monitoring of authority and insurance changes.
- Rate: pricing intelligence with tenant siloed rate memory built from what actually books, DAT benchmarks, and statistical anomaly flags on quotes.
- Tracker: load following through ELD fusion, late risk prediction, and drafted check calls.
- Vault: document extraction infrastructure. Every COI, rate confirmation, BOL, and POD classified, extracted, confidence scored, and filed.
- Scout: a voice agent that answers inbound carrier calls on a provisioned number, negotiates rates in the call, and runs live trust and fraud checks against the Dollar Traq database while the carrier is still on the line.
- Fleetra: the orchestrator. A broker types one plain request, Fleetra plans the work, fans it out to the specialist agents in parallel, and merges one sourced answer back.
Human in the loop by architecture
Approval is not a setting someone can switch off. Fleetra's approval gates are nodes in the orchestration graph itself, so nothing sends, pays, or disputes without a person tapping approve. Every session is checkpointed and replayable for audits.
DT Pay, the payments layer
Conditional payment release built on Stripe Connect. Funds are committed before pickup and released only when the proof of delivery is AI verified against the load and a human authorizes. Carriers are paid in about 90 seconds at a flat 1.2 percent fee. Every paid carrier was KYC verified at onboarding with a facial liveness check and CDL capture through the driver app we built in React Native.
The Impact
- Carrier vetting that took days of pulling records now happens in seconds, with a defensible score instead of a gut call.
- Fraud attempts such as chameleon carriers are flagged with evidence before a load ever moves.
- Brokers run vetting, pricing, tracking, paperwork, and payment from one system instead of stitching four tools together.
- First customers onboarded within weeks of launch, with the agent layer publicly live and a SOC 2 audit in progress.
- The platform is fully multi tenant and customizable per broker: settings, guardrails, and feature toggles for every agent.
Why This Matters for Your Business
This is the deepest expression of what ProMonkey builds: AI agents that do real operational work, grounded in owned data, with human approval gates engineered into the architecture rather than promised in a policy. If your industry runs on repetitive verification, document chasing, and phone calls, the same pattern applies.
Tech Stack
Python, FastAPI, LangGraph orchestration, Redis Streams event bus, PostgreSQL, Stripe Connect, Nylas email integration, Terminal ELD aggregator (Motive, Samsara, Geotab and more), React Native driver app, MCP based agent interfaces, Telnyx SMS authentication
FAQ
Did ProMonkey build the AI agents or integrate an existing product?
ProMonkey engineered the agents, the orchestration layer, the data warehouse, and the payment flows as custom software for Dollar Traq.
Does the AI act on its own?
No. Every payment, message, and dispute passes a human approval node that is part of the system architecture and cannot be bypassed.
Can this kind of system work outside freight?
Yes. The pattern of owned data, specialist agents, an orchestrator, and human approval gates applies to any operations heavy industry.
