TRUCK — Filling the Empty Passenger Seat

Ride along with an owner-operator for a day and you'll notice the most expensive thing in the cab is the empty seat. One truck, one driver, their own authority, 2,500–3,000 miles a week — and every decision that a trucking company spreads across a dispatcher, a back office, and a maintenance department lands on one person doing 65 mph. Which route clears a 13'6" bridge loaded at 13'8". Whether that new engine noise at 2 a.m., 300 miles from a trusted shop, means keep driving or stop now. Whether the load on the board is profit or a trap once you net out deadhead, fuel, and hours of service.

There are roughly 350,000 of these businesses in the US, most running one to five trucks. TRUCK is being built to fill their passenger seat — and the way it's being built is the case study.

The Product

TRUCK is a production-deployed web platform for commercial drivers, owner-operators, and small-fleet dispatchers: a purpose-built truck route planner with live turn-by-turn navigation, fleet dispatch with realtime positions and messaging, a load board with geospatial matching, corridor weather from NOAA, live CBP border wait times, a crowdsourced diesel-price map, a repair-shop directory, tax and per-diem tooling, a driver community — and Engine Shazam, an AI mechanical diagnosis that listens to your engine. Thirty-one routes, 111 components, 28 RLS-secured tables, ten edge functions: 36,305 measured lines of delivered code, live at truck.rlgix.com.

Consumer navigation is structurally wrong for a Class 8 truck — it routes under low bridges, ignores weight restrictions, knows nothing of hazmat class or border queues. The truck-specific incumbents are hardware-tethered ELD platforms with per-seat pricing that excludes the owner-operator entirely. TRUCK aims at exactly that excluded majority — which is why the build's first decision was the one everything else depends on.

The Engine: Own the Roads, Don't Rent Them

The obvious path was wrapping a commercial truck-routing API. It was rejected on arithmetic: those APIs bill per request, and a long-haul driver re-routing across a 2,000-mile run generates dozens of billable calls per trip — a cost structure that makes a freemium driver tier impossible before the first user signs up. So Rocket Now built the roads in-house.

The result is the project's crown jewel: a first-party routing engine — 1,794 lines in a single edge function — that constructs its road graph from OpenStreetMap data across three mirror endpoints with retry and backoff, solves it with A* search over a custom binary-heap priority queue, and prunes the search with a geodesic folding heuristic: a great-circle-geometry-biased admissible heuristic that keeps the frontier tractable at long-haul distances where naive heuristics fall apart. Constraints are enforced at the edge-traversal level — height, weight, axle load, hazmat class, tolls, ferries — so the illegal route is never generated, not merely flagged. Graph tiles cache into Postgres with a corridor pre-warming job that builds high-traffic freight corridors ahead of demand, and a hybrid "Baller Mode" stitches commercially-refined geometry into select segments while the first-party engine keeps the global solve.

When you own the roads, then the marginal cost of a route is compute plus a cached call — no per-route fee, ever. Add the operations layer running on free government data — NOAA forecasts resolved against actual route geometry, live border wait times with history, crowdsourced fuel prices that compound as a network asset — and three of six external integrations carry zero marginal cost. That's not frugality; it's the business model. The deterministic layer is free at the margin, which leaves exactly one variable cost in the whole platform: reasoning. And that's the premium tier.

The AI Layer: One Workload Live, Six Waiting for a Co-Driver

Engine Shazam is the proof the AI plumbing is real. A driver records the engine; an edge function submits a structured diagnostic prompt built on a 30-year diesel-mechanic persona covering nine failure signatures — belt squeal, bearing rumble, rod knock, and the rest — and enforces a typed five-field JSON contract with schema validation, severity coercion, confidence clamping, a conservative fallback when parsing fails, explicit rate-limit and quota handling, and persistence for longitudinal vehicle health. And deliberately, the model behind it is a swappable dependency on an OpenAI-compatible contract: repointing it is a base-URL, key, and model-id change against an already-proven pipeline.

That's the pattern for the six architected workloads in the registered Grok integration plan, and the roadmap reads like the passenger seat filling in stages: route reasoning, where the model consumes the engine's structured output and explains the why — "18 miles longer through I-84 because the direct route has a 13'6" bridge and you're loaded at 13'8"; you'll lose 22 minutes and save a strike"; the voice co-driver, the defining workload because a trucker's hands are on a wheel and eyes on a road — "find me diesel under $3.80 with a shower in the next hundred miles" — streaming, tool-calling, with a hard latency budget and zero interactions requiring a glance at the screen; load profitability verdicts on the board; fleet exception briefings for dispatchers; a tax and deduction assistant grounded in the driver's own records; and moderation keeping the community and fuel reports honest.

The commercial honesty is stated the way we always state it: TRUCK is pre-revenue by design — the team built the defensible routing core and the full operational surface before monetizing, all consumption figures in the plan are forecasts labeled as forecasts, and the closed beta exists precisely to replace projected numbers with measured ones before pricing is set against them. When the deterministic layer costs nothing at the margin, then the AI tier's unit economics are legible on day one — and the model partnership becomes a strategic choice rather than an incidental one.

The Results

A production-deployed trucking platform — 36,305 measured lines, 31 routes, 28 fully RLS-secured tables (82 policies), ten edge functions — live and serving real routes
A first-party routing engine with truck-legal constraint filtering at the edge level and zero per-route commercial fees — the architectural decision that makes a freemium driver tier economically possible
A live-operations layer on free public data: three ingestion pipelines (NOAA, US CBP, OSM), four realtime surfaces, $0 recurring data licensing
One production AI workload live with typed contracts, validation, and graceful degradation — and six more architected to deploy against an already-proven pipeline
A measurement-gated path to revenue: closed beta targets stated as targets, pricing set only against measured cost-per-driver

"[Quote pending — TRUCK stakeholder sign-off.]" — [Name, Title], TRUCK

Why It Worked: The Right Step Was Owning the Substrate

The Rocket Now principle, in diesel this time: a single step in the right direction is worth exponentially more than many steps in the wrong one. The wrong steps were renting — wrap the commercial routing API, ship faster, and inherit a cost structure that forecloses the freemium tier and a data granularity that can't tell an axle limit from a speed limit. The one right step was building the deterministic substrate first and owning it outright: the graph, the constraints, the caches, the government-data pipelines. Now every future capability gets cheaper — the AI workloads don't need to know trucking, because the substrate hands them structured truth to reason over. The hard, unglamorous data work is done. When the roads are yours, then the co-driver you hire only has to think.

Building in an industry the software world priced itself out of? Own the substrate. We'll pour it with you. → RocketNow.com

Editor's note — before publishing: