HYPR Fleet OS — EV Fleet Operating System
12M gig workers. 60% churn in 90 days. ₹7k/mo invisible tax. No portable identity. HYPR needed one OS to onboard, house, gear, and credit-score riders — not 7 disconnected tools.
I designed and built a full-stack EV fleet operating system spanning fleet management, rider onboarding, KYC, vehicle workflows, Data Harness, AI automation, WhatsApp/WhatsApp API operations, document processing, Cloudflare Queues and R2. Live at fleet.example.com + rider.example.com + wa.example.com.
Architecture
┌──────────────────────┐
│ Fleet / Rider │
│ Web Apps │
└──────────┬───────────┘
▼
┌──────────────────────┐
│ Fleet OS │
│ Product + Workflows │
└──────────┬───────────┘
┌─────────────┼─────────────┐
▼ ▼ ▼
┌────────────┐ ┌────────────┐ ┌─────────────┐
│ PostgreSQL │ │ Cloudflare │ │ WhatsApp API │
│ PostgreSQL database │ │ Workers │ │ WhatsApp │
└────────────┘ │ Queues/R2 │ └──────┬──────┘
└─────┬──────┘ │
│ ▼
│ ┌──────────────┐
│ │ WhatsApp │
│ │ Intelligence │
│ └──────────────┘
▼
┌───────────────┐
│ AI / Data │
│ Intelligence │
└───────┬───────┘
▼
┌───────────────┐
│ Recommendations│
│ + Approvals │
└───────────────┘What I built
Fleet Management
Fleet dashboard, vehicles, rider management, operational status, utilization. Command Center [private hub] with real-time fleet visibility.
Rider Operations
Onboarding, profile, KYC, document workflows, vehicle assignment, operational status. 7-step wizard writing to PostgreSQL database.
Data Harness
Company → Fleet → Vehicle → Rider. Canonical data model with RLS, scope, classification, validation, permissions, AI policy, audit.
WhatsApp Operations
WhatsApp → WhatsApp API → Fleet operations → AI intelligence. Single transport [private gateway], wa.example.com gateway.
AI Operations
Operational data → AI context → signal detection → recommendation → approval → action. human-in-the-loop approval for HIGH risk, auto-reply via [private].
Document / OCR Processing
Cloudflare Workers + Queues + R2 + OCR. Idempotent job processing hash(messageId+attachmentId), Evidence 96% confidence.
Technical decisions
Engineering judgment — why I chose each technology and what it solves.
Why Cloudflare Workers?
Edge execution, lightweight APIs and integration with the existing data layer. Queues for async processing, R2 for document storage, Containers for warm DR. One ecosystem, not five vendor contracts.
Why LLM inference?
Low-latency inference for real-time WhatsApp classification. Sub-second response times for operational AI that needs to act fast on incoming messages.
Why Supabase?
Structured operational data, authentication and relational querying. RLS for scoped access. One database for Fleet, Riders, Vehicles, Bookings, Evidence — not scattered across services.
Why WhatsApp API + WhatsApp?
WhatsApp is how riders actually communicate. WhatsApp API gives programmatic access. Single transport [private gateway], wa.example.com gateway. No per-rider sessions.
Technologies
Live demos
fleet.example.com
Fleet OS dashboard — Command Center, Fleet, Riders, Bookings, Service, Charging, Revenue, AI, Approvals, Ledger.
Visit live →rider.example.com
Rider app — Choose Bike, Mobile + Hub, Fast KYC camera, Confirm & Book. 7-step onboarding wizard.
Visit live →wa.example.com
WhatsApp gateway — WhatsApp API gateway, reverse proxy, health check endpoint.
Health check →Private repo
vridhilabs/fleet-hyper — Private repository, hiring-manager access available on request.
Contact for access →
Evidence
WhatsApp → human-in-the-loop approval
[private] → [private] → [private] → [private hub]. Text inline triggers human-in-the-loop approval audit event + auto-reply via [private]. Images publish document processing job to Cloudflare Queue.
KYC → R2 Storage
local validation (Indian ID validation Verhoeff) + mock KYC adapter → format validated. R2 [private bucket]/kyc/{riderId}/{type}/{jobId}.jpg private, presigned URL with 5min expiry.
Cloudflare Infrastructure
Queue ocr-job [private queue] producers:1 consumers:1. Worker [private worker] POST /ocr → 202 queued. Container [private container] warm instances:0.
Kill Test (Planned)
primary server primary server failure → human-in-the-loop approval ✓. Kill primary server → Cloudflare warm DR → 834 → human-in-the-loop approval. Restore primary server → no duplicate.
Skills demonstrated
Product
Fleet Ledger data model, rider onboarding UX, Fleet OS dashboard, approval workflows
Engineering
Next.js, TypeScript, Supabase RLS, Cloudflare R2/Queues/Workers/Containers, WhatsApp API, LLM inference
AI / Automation
Data Harness, AI Policy, human-in-the-loop approval, idempotent document processing job processing, WhatsApp intelligence
Infrastructure
primary server → Cloudflare DR, warm standby, DNS only, reverse proxy, firewall