The owner today
Pen, paper and memory
Credit given at the gate, diesel slips in a drawer, vehicle papers renewed only when a fine arrives. The business runs on what one person can remember.
A proposal from EZiL · September 2026
Every morning, one message the owner can trust.
Subah turns bills, voice notes and phone calls into a ledger that keeps itself, then hands it to the owner of a rice mill, crusher or trading yard as a single WhatsApp message before the day begins.
- For
- Rice mills · Crushers · Scrap · Transport
- Lives in
- WhatsApp, where owners already read
- Kept true by
- AI capture + a human verification desk
‹
SubahBusiness account
Today
सुप्रभात, Rameshji 🙏
Your morning at Shree Ganga Rice Mill
💰 To collect today
- Mahesh Traders₹1,84,00038 days
- Gupta Agro₹92,50021 days
- R.K. Stores₹41,2009 days
📅 Coming due
- Road tax · CG04 AB 2211 · 18 Oct
- Insurance · CG04 KD 7781 · 20 Oct
⛽ Fuel & vehicles
Diesel yesterday 240 L across 3 trucks. CG04 AB 2211 used 18% more than its weekly average.
📰 Market
Paddy procurement window extended in Chhattisgarh till 15 Nov.
Sample message · Figures are illustrative
सुबह Subah means "morning": the one moment this whole system exists to protect. It's a working name, and yours to rename.
What we understood
You aren't really asking for an ERP.
You're asking for trust, delivered daily.
What the owner actually needs
Three answers in two minutes
Who owes me? What's due? What did I spend? Answered before the mill opens, in the app they already check fifty times a day.
What the company becomes
The daily habit of thousands
Once thousands of owners open the same trusted message every morning, that trust becomes a platform for everything that comes after it.
The strategic bet isn't "build every feature." It's to own the daily trust moment first, then build everything else on top of it.
Our reading of your blueprint, and the principle this proposal is organised around
The product, in one screen
Four answers, one message, 6:00 AM.
The owner never logs into anything. Each block in the message is backed by its own engine, schedule and verification rule.
The rule behind it
A party appears when an unpaid balance crosses its age threshold. The list is sorted by amount × days outstanding and capped at five, so the message stays readable.
Source of truth
Ledger entries · verified by the field desk
How it works
From a crumpled slip to a ledger line, with nobody typing.
Pick an input and follow it through the four stages every piece of information passes through.
-
1Capture
-
2Understand
-
3Verify
-
4Ledger
The architecture underneath
InputEverything the owner already produces
- Text
- Bill photos
- Voice notes
- Field desk calls
- CCTV · later
- Weighbridge · later
UnderstandAI reads, a rules engine decides
- OCR · Vision / Textract
- Speech-to-text · Indic languages
- Claude · structuring & anomaly checks
- Rules engine
- Confidence scoring
Ledger coreOne vault per business
- Parties
- Ledger entries
- Vehicles & obligations
- Employees
- Field desk
- Audit trail
DeliverWhere the owner already is
- WhatsApp · daily 6 AM
- SMS fallback
- Owner & supervisor dashboard
- Owner app · later
The stack we recommend, and why
| Layer | Our choice | Why this, for this product |
|---|---|---|
| Owner channel | WhatsApp Business Platform via a BSP | Approved utility templates reach the inbox reliably. Delivery receipts tell us who read their morning, and SMS takes over when WhatsApp fails. |
| API & workers | Python · FastAPI | The AI side of the product (OCR, speech, Claude, anomaly rules) is native to Python. One language from ingestion to ledger. |
| Database | PostgreSQL + row-level security | Financial data needs constraints and transactions. Row-level security on business_id makes isolation a database guarantee, not a coding habit. |
| AI structuring | Claude API | Turns messy Hinglish transcripts and blurred bills into typed entries, and explains each entry in plain language when a human needs to check it. |
| Speech & OCR | Whisper-class STT · Google Vision / Textract | Tuned for Hindi and regional languages, handwritten slips and thermal-printer receipts. |
| Files & hosting | AWS Mumbai region · S3 | Bill photos and voice notes stay in India, in line with the DPDP Act. It scales from a pilot to thousands of owners without a rebuild. |
| Dashboard | React + Tailwind | One web app serves owner, supervisor, BRM and leadership views, each with its own permissions. |
| Owner app | Flutter (Phase 3) | One codebase for Android-first owners. It ships after the WhatsApp habit is proven, not before. |
The data model
Your blueprint, made audit-proof.
We keep every entity you defined, and strengthen the design where money is involved. Every table carries a business_id, so no owner ever sees another owner's data.
Business & owner
Type, WhatsApp number, plan and status. Drives which alerts and modules each owner gets.
- business_type
- whatsapp_number
- plan_status
Party refined
Creditors and debtors in one table, because the same trader often buys from and sells to the mill.
- name · phone
- balance (derived)
- last_payment_date
Ledger entry
Append-only. Every entry keeps its original photo or voice note and records who verified it.
- source_type · raw_source_ref
- type · amount · party
- confidence · verified_by
Vehicle & obligations refined
Tax, insurance and permit become rows, not columns, so fitness, loans or GST filing can be added later without a redesign.
- registration_number
- obligation · due_date
- pending · paid · overdue
Employee & attendance
Starts with manual attendance. The source switches to CCTV AI in a later phase with no schema change.
- name · role
- attendance_source
Field desk
Agents, their assigned owners, and every call with its checklist, so data quality becomes something you can measure.
- assigned_business_ids
- call_log · checklist_result
Alert log added
Every morning message: what it said, whether it was delivered and whether it was read. This is the heartbeat of the business.
- sent · delivered · read
- fallback_channel
Where we'd strengthen the blueprint
- Balances are derived, never overwritten. An owner disputing a number gets the full trail of entries behind it, with the original bill.
- Confidence decides the route. High-confidence entries post straight away; low-confidence ones go to the field desk before the owner sees them.
- Isolation lives in the database. Row-level security enforces
business_idon every query, so one owner's data can't leak to another even if the application code has a bug.
The human layer
Owners won't pay for wrong numbers.
So people keep the numbers right.
Five levels, and we build the software each one works in. Every level gets exactly the view it needs and nothing more.
-
L5
Managing Director
Sets vision, expansion and what gets built next.
Works in
Company cockpitActive owners, morning read rate, churn, data accuracy by region -
L4
Regional / Operations Manager
Owns a state: hiring, growth targets, recurring issues.
Works in
Regional boardTeam capacity, owner growth and escalations that keep coming back -
L3
Business Relationship Manager
Owns relationships: renewals, upgrades, disputes, feedback.
Works in
Portfolio viewOwner health scores, at-risk alerts and upgrade readiness -
L2
Supervisor
Checks call completion and data quality, and escalates stuck owners.
Works in
Quality queueMissed calls, unverified entries and agent accuracy -
L1
Telecaller / Field agent
Maintains a rolling book of owners by phone and visit.
Works in
Field deskToday's call list, a guided checklist and one-tap entry confirmation
What we protect
Five promises built into the code.
i
No number without a source
Every figure in the morning message links back to a bill, a voice note or a verified call.
ii
A person confirms what AI isn't sure of
Uncertain entries wait for the field desk. The owner only ever sees numbers someone stands behind.
iii
One business, one vault
Isolation is enforced by the database itself. Data stays in India and access is logged.
iv
The morning feed stays sacred
Brand content, when it comes, lives in the app, never inside the trusted alert.
v
It fails loudly, then falls back
An undelivered WhatsApp triggers SMS and a flag for the desk. No owner silently misses a morning.
Roadmap
Earn the habit first. Then widen it.
Each phase ships something an owner uses every day, and each one is gated on proof from the phase before it.
Phase 1 · Pilot
The first morning
- Ledger core with per-business isolation
- Text and voice-note capture (Hindi + English)
- Daily WhatsApp: creditors and due dates
- Vehicle obligations with repeating reminders
- Field desk console for telecallers
Moves on when pilot owners read the message daily and zero balances are disputed.
Phase 2 · Scale capture
Nothing gets typed
- Bill and receipt OCR with confidence routing
- Fuel and vehicle intelligence
- Keyword-matched market news
- Owner and supervisor dashboard
- Supervisor queue and BRM portfolio
Moves on when most entries arrive by photo or voice and are verified the same day.
Phase 3 · See the yard
The mill, observed
- CCTV-based attendance
- Weighbridge readings into the ledger
- Owner app on Android and iOS
- Regional board and company cockpit
- In-app brand content surface
Moves on when owners ask for it, which is exactly what the BRM layer is there to hear.
How we'll know it's working
We measure the pilot by the owner's morning.
Daily
Owners open the 6 AM message on most days without being reminded
Same day
Captured entries are verified by the desk before the next morning's message
Zero
Balances disputed because the system got them wrong
Before
Every vehicle obligation is flagged before a penalty, not after
Next step
Let's build the first morning.
We propose starting with a small group of pilot owners: one business type, one region and one field desk. Within the first phase, every one of them wakes up to a message they believe.
- Discovery call. We choose the pilot owners together and sit with two of them for a day.
- Message design. We lock the morning message format and language with real data.
- Build and launch. Phase 1 goes live with the field desk trained alongside it.
Write to us at contact@ezil.work · open in mail