Aleph Hackathon

What it does

Choqui is a WhatsApp-based assistant for car accident insurance claims, paired with a management dashboard for insurers. When someone crashes, they message Choqui instead of calling a call center. The bot validates the license plate against an active policy, checks for injuries first, gathers the account of what happened, and sends a single link where the user confirms the location on a map and uploads all required documentation — all in one session, no back-and-forth between WhatsApp and the browser.

How it works

1. Policy validation — the bot checks the license plate against an active policy before creating a case.
2. Injuries first — if anyone is hurt, the bot redirects to emergency services and flags the case for human review.
3. Conversational account — the bot gathers what happened, how, and when, through natural conversation.
4. One link, full checklist — a single mini-app link covers location confirmation (interactive map) and 11 required documents. Every photo is taken live with the camera (never from the gallery), geotagged, and hashed to detect duplicates.
5. Automatic handoff — once the checklist is complete, the system automatically notifies the bot to continue the conversation — no user action required beyond finishing the upload.
6. Fraud analysis — a hybrid engine combines explainable rules (unverified photos, inconsistent GPS location, duplicate images, repeated license plates across claims) with a second AI opinion that can adjust the score within a bounded range, so operators see why a case is flagged, not a black box.
7. Operator dashboard — real-time claims inbox, a queue of new cases, and a fraud diagnostics view ranked by risk.

Why it's different

• Real verification, not just a chatbot. Forced camera capture + live geolocation + file hashing make photo fraud significantly harder.
• A structured AI agent, not a freeform chatbot. The agent converses naturally but writes every state change to the database through a scoped tool — nothing is invented or lost.
• Explainable fraud detection. Rule-based scoring plus a bounded AI second opinion, never an unexplainable black-box number.

Business model

B2B2C: insurers pay a monthly license plus a per-claim fee. The end user (the driver) uses it for free — replacing the call center they already pay for today, but faster and with verified data.

Tech stack

• n8n orchestrates the WhatsApp bot, a fraud-analysis workflow triggered by Supabase Database Webhooks, and OpenAI (text, voice transcription, vision, and a tool-using agent)
• Supabase for Postgres, Auth, Storage, and Realtime updates
• Twilio for the WhatsApp channel
• Next.js on Vercel for the operator dashboard and the mobile-first capture mini-app (map + document wizard)
• Redis for message debouncing and structured conversation state (awaiting license plate, awaiting continuation decision on a pending claim)