chitti-scanner

name

chitti-scanner

description

Chitti Product Scanner — capture-or-type analyser for Indian families. User uploads a photo (or types out the label, when camera fails) of a product label, medicine strip, bill, or document; Chitti returns product type (food / medicine / legal_doc / bill / mrp / insurance / other), key findings, warnings, and savings suggestions. Cross-links to Chitti MedUPI for Jan-Aushadhi alternatives, to Chitti UPI Fraud Guard for insurance-premium safety checks, and to Chitti Vaani for read-aloud results. Type-specific legal disclaimer appended to every response. Use on chitti_scanner.html or for any "is this expired", "is this MRP correct", "what does this medicine label say", or "explain this bill / contract" question.

Chitti Product Scanner — top-level skill

This skill describes the Chitti Product Scanner product as a whole. It loads when the user asks about scanning a label, reading a medicine strip, decoding a bill, summarising a contract / policy, or checking food labels for FSSAI info.

Repo layout

chitti-scanner/
├── frontend/index.html         mirror of workspace-root chitti_scanner.html
└── backend/
    ├── main.py                 Flask app · CORS · /api/scanner/* registration
    ├── config.py               settings (DEEPSEEK_API_KEY · vision model · CORS · MEDUPI_API_BASE)
    ├── requirements.txt        flask · flask-cors · gunicorn · httpx
    ├── runtime.txt             Python 3.11
    ├── render.yaml             Railway Blueprint
    ├── services/
    │   └── scanner_service.py  DeepSeek wrapper + CHITTI_SCANNER_PROMPT + per-type legal lines + cross-product links
    └── routes/
        └── scanner.py          Blueprint /api/scanner/*

Endpoint surface (Flask Blueprint, prefix /api/scanner)

Method	Path	Purpose
POST	/analyze	Either multipart image (field image, language form field) OR JSON {text, language?}
POST	/analyze/text	JSON-only path — {text, language?}
GET	/health	DeepSeek + vision-model status

Response shape:

{
  "ok": true,
  "type": "food | medicine | legal_doc | bill | mrp | insurance | other",
  "summary": "...",
  "facts":   { "brand": "...", "expiry": "..." },
  "key_findings": ["...", "..."],
  "warnings":     ["..."],
  "savings":      ["..."],
  "speak_hi": "<read-aloud line>",
  "speak_en": "<read-aloud line>",
  "legal_disclaimer": "<type-specific Hinglish line>",
  "cross_links": [{ "product": "medupi" | "upi_guard" | "vaani" | "consumer_helpline", ... }]
}

Legal disclaimers (type → line)

Type	Hinglish line
food	Yeh FSSAI label ki information hai. Dietary advice ke liye nutritionist se milo.
medicine	Yeh sirf label ki information hai. Doctor se confirm karo pehle.
legal_doc	Yeh AI summary hai. Final decision apne aap lo ya vakeel se lo.
bill / mrp	Agar overcharging hai toh consumer helpline 1800-11-4000 pe call karo.
insurance	Premium pay karne se pehle UPI Fraud Guard mein check kar lo.

Consent gating

localStorage.chitti_scanner_consent_given. T&C modal sections:

1. What Scanner is (and is not — no diagnosis, no legal verdict)
2. Legal Compliance (FSSAI · Consumer Protection Act 2019 · DPDP Act 2023)
3. Permissions Required (Camera access — only when you tap the camera button)
4. Data Privacy (images not stored on server; processed in-memory)
5. Emergency and Liability (consumer helpline 1800-11-4000 · medical → doctor)
6. Grievance Redressal ([e-mail])

Cross-product hooks (built-in)

• Medicine detected → cross_links returns a medupi_lookup hand-off; the frontend calls ${MEDUPI_API_BASE}/api/medupi/medicine/<query> and renders the Jan-Aushadhi panel inline.
• Insurance detected → frontend offers a deep-link to chitti_upi.html with the policy summary pre-filled.
• Food detected (or any time the user wants) → "Have Chitti read this aloud" deep-links to chitti_vaani.html?from=scanner with the result handed off via sessionStorage.
• Bill / MRP overcharging → tel-link to consumer helpline 1800-11-4000.

Live URLs

• Frontend: https://sahayai.in/chitti_scanner.html
• Backend: https://chitti-scanner-api-production.up.railway.app (planned — render.yaml ready)