<?xml version="1.0" encoding="UTF-8"?><oembed><type>video</type><version>1.0</version><html>&lt;iframe src=&quot;https://www.loom.com/embed/7a675a1918bf4233bd44c8e049f84c97&quot; frameborder=&quot;0&quot; width=&quot;1216&quot; height=&quot;912&quot; webkitallowfullscreen mozallowfullscreen allowfullscreen&gt;&lt;/iframe&gt;</html><height>912</height><width>1216</width><provider_name>Loom</provider_name><provider_url>https://www.loom.com</provider_url><thumbnail_height>912</thumbnail_height><thumbnail_width>1216</thumbnail_width><thumbnail_url>https://cdn.loom.com/sessions/thumbnails/7a675a1918bf4233bd44c8e049f84c97-2a64367e854e585f.gif</thumbnail_url><duration>269.071</duration><title>Dukkan Saati, Offline AI for Kirana Shops</title><description>This Loom demonstrates Dukkan Saati, an AI shop companion that automatically maintains kirana ledgers using voice and bill photos. The system runs on small open-source models: speech-to-text captures Hindi requests, and a vision model reads even messy handwritten supplier bills via OCR, with an editable tape for corrections. An agent acts like a clerk with a fixed set of tasks, reading the books back and waiting for “yes” before updating, and can reply in Hindi and even generate morning briefs. In a live example, it reported total sales of 6,527 rupees from 29 transactions and 104 units sold, listed best-selling items, and showed top debtors, credit reminders via WhatsApp, stock and expiring inventory, and optional analysis for slow movers.</description></oembed>