Inspiration

A dairy owner handles two accounts every day: milk bought from farmers and milk delivered to customers. A changed rate, a missed entry or a partial payment can make a paper register difficult to reconcile. Doodh Hisaab brings those records into one place, with a phone-friendly interface and clear amounts in rupees.

What it does

Doodh Hisaab records milk collections, confirmed deliveries, customer Udhaar, farmer payables, payments and stock movements. Each transaction keeps its original rate. Separate business accounts, server-side permissions and audit records protect the bookkeeping.

The Smart Assistant accepts supported English, Hindi and Hinglish requests. It can show daily entries, monthly summaries and outstanding balances. A request such as “Add 10 litres of milk for Demo Rahul” first produces a preview using that saved customer's rate. A separate confirmation is required before the delivery is saved.

The Alexa-style conversation view is a working browser simulation using the same command processor. It is not connected to an Echo device, Amazon Alexa+ or an MCP server. This submission uses the simulated-experience alternative in the Alexa+ track.

A small example that can be checked

The demonstration uses fictional records only. Ten litres at ₹60 per litre produce a ₹600 delivery. The preview creates no financial record. After explicit confirmation, a separate ₹200 cash receipt leaves ₹400 outstanding. The customer statement and downloaded PDF show the same result. The 2:28 demonstration presents actual application snapshots with fictional records, Hindi AI narration and English translation captions.

Retrying the same confirmed preview returns its existing receipt rather than creating another sale. Changed rates, expired previews, ambiguous customer names and unauthorized requests are checked on the server.

How it was built

The interface uses React, TypeScript, Lucide icons and restrained Motion animations. Shared JavaScript backend functions run locally with SQLite and on Cloudflare Pages/Workers with D1. Money is calculated in integer paise and milk quantity in thousandths of a litre, with India business dates.

The command processor is deterministic and does not require a paid model API. A separate Cloudflare Workers AI help mode answers product questions within a configured allowance. Generated text is never the authority for a customer balance or a financial write.

Work completed during development

The existing dairy application was preserved while the conversational query, preview and confirmation workflow was added. The upgrade also reduced public-page CSS, improved touch targets and prepared setup instructions, tests and synthetic demonstration assets. Git history keeps the earlier application and later changes distinguishable.

Challenges and verification

The most important challenge was making a convenient command safe to confirm. A preview can become stale if a customer rate, permission or subscription changes. Confirmation therefore checks the current records again and uses the application's existing transaction and stock path.

The source passed 275 automated tests, including 17 Smart Assistant tests, plus 27 release checks; six SEO tests overlap between those groups. The final QA candidate passed 84 HTTP checks. An earlier QA release passed 96 workflow checks; the final candidate uses byte-identical Worker code. These results do not replace physical-device testing or unverified payment-provider scenarios.

What comes next

The next priorities are official per-dairy WhatsApp onboarding, hosted backup-and-restore rehearsal and further Android testing. A funded live payment capture, refund and settlement remain unverified. The project was developed with coding assistance and review; its calculations, permissions and demonstrated results are checked independently of model-generated answers.

Share this project:

Updates

Submission history