Safari OS

Problem

Safari operators run bookings, pricing, supplier coordination, and client communication across spreadsheets, WhatsApp, and email, with no single system tying it together.

Approach

Building an end-to-end ERP for a real safari company covering bookings, a costing engine, CRM, itinerary builder, invoicing, and supplier management — over 30 Prisma models — with an AI layer (Claude, with Mistral fallback) for conversion scoring and anomaly detection.

Tech highlights

Next.js 16 with Prisma and PostgreSQL; JWT sessions with OTP and three-role RBAC (Admin/Ops/Guide); Twilio WhatsApp integration across 14 message types with real-time delivery receipts; PDF/DOCX document generation via PDFKit and an AI content studio generating marketing copy in 5 languages across 6 channels; Vercel Cron for scheduled jobs.

Results

In active development for a real safari operator — full write-up and demo coming once the platform is live.

Challenges

Keeping the costing engine accurate across multi-currency quotes, seasonal supplier rates, and per-person markups while modeling over 30 interrelated Prisma entities; and making sure the AI conversion-scoring and anomaly-detection layer degraded gracefully to the Mistral fallback so a slow or unavailable model could never block a booking or a WhatsApp reply.