kirana-khata

Voice ledger for kirana shops: track udhaar, record payments and draft reminders in Hinglish. An Alexa+ MCP server that even knows "chhota bhaiya" by nickname.

Comment

Inspiration

India has millions of kirana stores, and almost all of them run on udhaar (credit). The shopkeeper's ledger is a paper notebook, and his hands are usually busy weighing, packing, or handing over change. Writing "Ramesh owes 200" in a notebook mid-rush, or remembering who owes the most at the end of the month, is slow and error-prone.

Voice is the natural interface for this. A shopkeeper can just say it. So we built Kirana Khata, a voice-first udhaar ledger for Alexa+ that understands the way shopkeepers actually talk: Hinglish, with nicknames.

What it does

Kirana Khata lets a shopkeeper manage credit entirely by voice:

  • "Ramesh ke naam 200 ka udhaar likho" records a new credit entry.
  • "Sharma ji ne 500 diye" records a payment.
  • "Chhota bhaiya ne 500 diye" works too. Customers are known by nicknames, and Kirana Khata maps them to the right account.
  • "Aaj kitna udhaar baaki hai?" gives the total outstanding balance.
  • "Sabse zyada paisa kis se lena hai?" lists the top defaulters.
  • It can also draft a polite reminder message for a customer with pending dues, ready to send.

How we built it

Kirana Khata is an MCP server for the Alexa+ track. Alexa+ handles the speech and understanding; our server exposes the ledger as tools that actually write data (not just answer questions):

  • add_udhaar
  • record_payment
  • get_balance
  • top_defaulters
  • draft_reminder

Data lives in SQLite, so there are no external API dependencies and the system is simple and reliable. Nickname support is handled through an alias layer, so several names can resolve to the same customer account. [Add your language/framework, e.g. Python + MCP SDK, and how you deployed or tested it, e.g. real Alexa+ or a simulated Alexa+ experience.]

Challenges we ran into

  • Hinglish understanding: the same request can be phrased many ways ("likho", "daalo", "chadha do"). [Describe how you handled this.]
  • Nicknames and ambiguity: [Describe how you resolve a nickname like "chhota bhaiya" to one customer, and what happens when two customers share a name.]
  • Trust in financial data: since the tools change real balances, [describe any confirmation or validation you added].
  • [Add 1 or 2 real challenges you faced, e.g. time limits or tool-calling issues.]

Accomplishments that we're proud of

  • A real problem for a huge, underserved user base, solved with tools that take real actions.
  • Natural Hinglish voice interaction, including nickname-based entries.
  • A working end-to-end demo with no external dependencies.

What we learned

  • Designing MCP tools that are small, clear, and safe to call by voice.
  • Why voice-first design matters when users' hands are busy and typing isn't an option.
  • [Add your own learning, including about the Alexa+ and MCP developer experience.]

What's next for Kirana Khata

  • Hindi, Marathi, Gujarati, and other regional-language support.
  • Sending reminders directly over WhatsApp or SMS.
  • Daily and weekly summaries, and due-date tracking.
  • Multi-shop and multi-staff support, with backups.

Built with

Alexa+, MCP (Model Context Protocol), SQLite, [your language/framework]

Built With

Share this project:

Updates

Submission history