Compass Solutions
GPS fleet-tracking SaaS with AI-augmented operations
Production fleet-management and GPS-tracking platform that helps businesses monitor vehicles, manage operational data, and act on real-time location-based workflows. AI-augmented modules surface insights from telemetry; the Compass Assistant lets operators ask fleet questions in plain language and get back structured answers with deep-links into the relevant pages.
Live-map view — every active vehicle on the customer fleet, updated in real time via WebSocket telemetry stream.
10+
production modules owned
Real-time
GPS + telemetry tracking
AI-powered
driver / fuel / maintenance
Multi-tenant
fleet & organization scoping
Overview
Fleet operators don't need more dashboards. They need answers.
Compass is a GPS fleet-tracking platform built for businesses that need to monitor vehicles in real time, manage operational data, and act on location-based workflows — dispatching, geofencing, fuel monitoring, driver behavior, and predictive maintenance.
What sets it apart from a generic fleet-tracking SaaS is the AI layer. Compass ships AI-augmented modules for drivers, fuel, and maintenance that surface insights from raw telemetry without engineers having to query Postgres directly. The Compass Assistant takes that further: operators ask plain French/English questions like “which vehicles drove the most yesterday?” and get a structured answer with deep-links — no SQL required.
I work across the full stack on this product — web app, backend API, PostgreSQL data layer, generated API contracts, deployment workflows — and own the integration layer between real-time location data and the operational dashboards.
Modules
Six core modules I work across
Each module spans frontend dashboards, backend services, the PostgreSQL data layer, and the deployment workflow — end-to-end ownership across the stack.
Live Map
Real-time vehicle positions on an interactive map with cluster markers, route history, and per-vehicle telemetry overlays. Built with Mapbox GL + WebSocket streams from the GPS backend so fleet managers see vehicle state without refreshing.
Next.js · Mapbox GL · WebSockets · PostgreSQL
Vehicles & Devices
Per-vehicle dashboards with status (moving / stopped / offline), assignment to drivers, geofences, telemetry history, and document attachments. CRUD across the entire fleet with bulk operations and CSV export.
Next.js · TanStack Query · PostgreSQL · React Hook Form
Alerts & Geofences
Configurable alert rules — speed thresholds, geofence enter/exit, harsh braking, idle time. 99+ alert volume per day across the fleet with severity classification, assignment workflows, and acknowledgement tracking.
PostgreSQL · Cron jobs · Notification fanout · Real-time WebSocket push
AI Modules — Drivers · Fuel · Maintenance
LLM-augmented modules that surface insights from raw telemetry: driver-behavior scoring, fuel-anomaly detection (per-trip vs fleet baseline), and predictive-maintenance recommendations from mileage + sensor patterns.
LLM agents · Telemetry analytics · PostgreSQL aggregations
Compass Assistant
Natural-language fleet operator — engineers ask questions in plain French/English ("which vehicles drove the most yesterday?", "show me alerts on vehicle 12 this week") and the assistant returns structured answers with deep-links into the relevant pages.
LLM · Tool calling · Custom Compass tool registry
Reports & Documents
Custom report builder with date ranges, vehicle filters, and KPI selectors. PDF/Excel export. Document wallet for vehicle paperwork (insurance, registration, inspection) with expiry-date alerts.
PDF generation · Cloud storage · Cron jobs
Vehicles & Devices
One place to see and manage every vehicle
Per-vehicle dashboards showing status (moving / stopped / offline), driver assignment, telemetry history, geofence memberships, document expiries, and alert history. Bulk operations + CSV export for ops teams.
Compass Assistant
Natural language, grounded answers
LLM tool-calling agent purpose-built for fleet operations. It resolves plain-language questions into the right tool calls (vehicles, alerts, geofences, fuel, drivers), executes them against the live database, and renders a structured answer with deep-links into the relevant pages — no hallucinations because every answer is grounded in tool output.
Engineering Challenges
The hard parts I worked on
Real-time at fleet scale without melting the database
Solution · Telemetry streams arrive at high cadence — naive 'just upsert into Postgres' would crush write IOPS. Built a buffered ingestion layer that batches inserts every N ms, separate hot-cache for the live-map view, and aggregated rollups for historical queries so the map stays sub-200ms responsive while the backend isn't constantly contending on row locks.
Outcome · Live-map updates feel instant; historical queries don't compete with the ingest pipeline.
Multi-tenant access across vehicles, drivers, organizations
Solution · Each user belongs to one or more organizations; each organization owns vehicles, drivers, geofences, and reports. Built a fleet-scoped access layer at the query level so that no cross-tenant leakage is possible, plus a clean RBAC model surfaced through a Team page where org admins manage roles.
Outcome · Organizations can grow into thousands of vehicles without leaking data across tenants.
AI assistant that actually answers fleet questions
Solution · Built a tool-calling layer specific to fleet operations: vehicles, alerts, geofences, fuel queries, driver lookups. The assistant translates plain-language questions into the right tool calls, fetches the data, and renders an inline answer with deep-links — no hallucinations because it's grounded in the database.
Outcome · Fleet managers get answers in 5 seconds instead of clicking through 6 dashboards.
Engineering-friction reduction across the codebase
Solution · Introduced reusable UI patterns, structured service logic, database migrations, seed scripts, type-checking, and build-quality improvements across the platform so future contributors ship faster with fewer regressions.
Outcome · New modules now reuse 60-70% of existing primitives instead of reinventing forms, RBAC, and notifications.
The Stack
How it's built end-to-end
Frontend
- Next.js 14 (App Router)
- TypeScript strict mode
- Tailwind CSS · Radix UI
- Mapbox GL JS
- TanStack Query
- React Hook Form · Zod
Backend
- Node.js · TypeScript
- REST + tRPC
- WebSocket server (real-time)
- Cron jobs (alerts, reports)
Data
- PostgreSQL
- Prisma ORM
- Telemetry buffering layer
- Multi-tenant row-level scoping
- Aggregated rollups for analytics
AI & Infra
- LLM tool-calling agent
- Custom Compass tool registry
- Cloud object storage (documents)
- Vercel-style deployment workflow
What I Deliver
Day-to-day at Compass
- 01
Built and refined dashboard, vehicle, driver, map, and fleet-management flows for clearer day-to-day tracking and improved operational visibility.
- 02
Worked across the full stack — web app, backend API, PostgreSQL data layer, generated API contracts, and deployment workflows — to strengthen end-to-end product delivery.
- 03
Reduced engineering friction by introducing reusable UI patterns, structured service logic, database migrations, seed scripts, type-checking, and build-quality improvements.
- 04
Owned the integration between real-time location data and operational dashboards so fleet managers can act on driver and vehicle state without leaving the app.
- 05
Shipped AI-augmented modules (Compass Assistant, AI driver/fuel/maintenance insights) that translate raw telemetry into operator-facing answers.
Building real-time SaaS for your team?
I work end-to-end on production systems with real-time data, multi-tenant access, and AI-augmented operations.