FleetLogicAI

FleetLogicAI — AI-Native Fleet Dispatch for the Other 90%

Inspiration

Most trucking software is built for enterprise fleets with IT departments and six-figure TMS budgets. But the majority of carriers in the U.S. run 5–15 trucks, managed by a single dispatcher juggling a spreadsheet, three load board tabs, and a phone that won't stop ringing.

The average small fleet dispatcher makes 40+ manual calls per day just to check driver availability. Every load assignment is a gut call. Deadhead miles bleed margin silently. Nobody is telling these dispatchers what things actually cost until the end of the month.

That felt like a real problem worth 24 hours.

What We Built

FleetLogicAI is an AI-native fleet operations assistant. A dispatcher uploads their fleet CSV, and the app immediately gives them a live dashboard showing every driver's status, location, and HOS availability pulled from the NavPro API. The AI then evaluates every driver against HOS compliance, proximity, truck capacity, and deadhead cost — and tells the dispatcher exactly who to assign and why, including a full cost breakdown per decision.

A dispatch chatbot handles follow-up questions like "what if Driver 3 calls out sick?" or "which routes are killing our margin?" in plain English.

How We Built It

Frontend: React + Tailwind CSS Data parsing: PapaParse for CSV ingestion Live fleet data: Trucker Path NavPro API — real driver locations, work status, and load assignments ELD integration: Mock ELD data in FMCSA format covering HOS, duty status logs, odometer, and violation flags AI layer: Claude API — receives the merged NavPro + ELD + CSV data as context and reasons over it to generate recommendations and answer dispatcher questions

The architecture is a two-layer system. NavPro is the eyes — it sees the fleet in real time. Claude is the brain — it reasons over everything and tells the dispatcher what to do.

Challenges

We came into this as newcomers to software development. Neither of us had shipped a production React app before.

The biggest early mistake was wiring the chatbot directly to NavPro and expecting it to respond conversationally. NavPro is a data API — it fetches, it doesn't think. We had to rebuild the architecture so NavPro feeds data to Claude, and Claude does the reasoning.

HOS compliance logic was harder than expected. Federal rules are genuinely complex — 11-hour drive limits, 14-hour shift windows, 70-hour cycles, 30-minute break requirements. Getting Claude to apply all of these correctly required careful system prompt engineering.

We also had to make the demo feel real. We seeded actual drivers into NavPro and built mock ELD data matching the FMCSA output file format so the demo reflects what production data would look like.

What We Learned

How the trucking industry actually works — loads, brokers, HOS regulations, deadhead economics. How to combine a data API and an LLM into a product that actually does something useful. That a great system prompt is as important as the code around it. That two people with no background can ship something real in 24 hours if the problem is clear enough.

What's Next

Live load board integration via the DAT Power API so FleetLogicAI can match drivers to available loads automatically. Real ELD provider connections through KeepTruckin and Samsara to replace mock data. Fuel price API integration for real-time cost calculations. Mobile view for dispatchers on the go.

Built With

• claude-api-(anthropic)
• cursor
• javascript
• navpro-api-(trucker-path)
• papaparse
• react
• tailwind-css
• vercel
• vite