Your data

A plain answer to where your data sits, how it is cleaned up and how it turns into reports. Also how a photo of an invoice becomes an entry in Tally.

Where your data lives

Most of what we build needs a copy of your data: customers, items, invoices, orders. That copy sits in one of two places, and you choose which.

The usual choice is a server in India. A server is a computer rented by the month in a data centre. Customer names, phone numbers and invoices stay inside the country, which matters for India's data protection law and for your own peace of mind. The server is rented in your name where possible, so the bill and the login are yours. If you would rather keep everything in the building, we can set it up on a machine in your office instead. It is a little more work to look after, and we will tell you honestly whether it suits you.

A server in India
Rented by the month in a data centre in Mumbai, Chennai or Bengaluru. Reachable from every branch. Backed up every night. The right choice for most businesses.
A machine in your office
Everything stays in the building and works without a fast internet line. You look after the power and the hardware; we look after the software.
A mix of both
Day-to-day work on the server, with a full copy pulled to a machine in your office every night. Some owners simply sleep better this way.

Whichever you pick, the data is yours. You can ask for a full copy at any time, in files that Excel and Tally can open, and you do not need a reason. See Keeping it safe for who can open what and how backups work.

How it is cleaned

Data from a running business is never tidy. The same customer is saved three ways. An item is spelt differently on the purchase side and the sales side. Codes changed two years ago and half the sheets still use the old ones. None of this is anyone's fault. It is what happens when five people type into four systems for ten years.

We clean it once, properly, before anything else is built. That means:

  • Duplicate customers and suppliers merged into one record each, with the old names kept as alternative spellings so nothing breaks.
  • Item names and units made consistent, so that "2 inch PVC pipe" and "PVC PIPE 2IN" become one item.
  • Codes matched across systems: Tally ledger names to the billing software, HSN codes to items, branch names to the ones on the GST portal.
  • Dates, amounts and phone numbers put in one format.

Then we keep it clean. The system checks new entries as they arrive. Anything that looks like a duplicate or an odd spelling goes on a short list for someone on your team to confirm each week. Five minutes on a Friday keeps the data usable for years. This is the first piece of Make your numbers work.

How reports are generated

Every report we build reads from the same cleaned data, and it is rebuilt every night. That one rule removes the most common argument in any office: the sales report says one number, the accounts report says another, and someone spends an afternoon finding out why.

On the first morning of the month, the monthly report is already there. Daily reports arrive on WhatsApp or email before the shop opens. Nobody exports, pastes or formats anything. If a figure looks wrong, the fix is made once, in the data, and every report agrees the next morning. What the reports can tell you is covered under Forecasts and analysis you can act on.

What we can read

Most of the data we need already exists. It is just sitting in the wrong shape. An export from Tally, a spreadsheet someone keeps by hand, a photo of a supplier's bill, a WhatsApp message with an order in it. Here is what we can read, where it comes from, and how well it works in practice.

Scroll sideways to see all the columns: From where and How well.

What we read, from where, and how well
WhatFrom whereHow well
Tally exportsLedgers, vouchers and stock, exported daily or read straight from Tally on your machine.Very well. The data is already in neat rows and columns, and once names are matched there are almost no errors.
Excel sheetsPrice lists, dispatch registers, attendance, anything a team keeps by hand.Well. We match the columns once. A change to the sheet's layout needs a quick re-check, which the system asks for.
Scanned invoicesPhotos or PDFs of supplier bills, sent on WhatsApp or email.Well for printed invoices, which mostly need no correction. Handwritten ones go to a person.
WhatsApp messagesOrders, questions and confirmations in English, Tamil or a mix of both.Well for orders with item names and quantities. Voice notes are written out first, then read.
EmailsPurchase orders, statements and attachments in a shared inbox.Well. The message is sorted and its attachment is read the same way as a scanned invoice.

How well is what we see in practice on real documents. We measure it on your own files in the first two weeks and show you the number.

From a photo to a Tally entry

Reading scanned invoices and Tally exports

A supplier's bill arrives as a photo on WhatsApp. This is what happens next, and where a person stays in charge.

Scroll sideways to see the whole route. The list below says the same thing in words.

How a photo of an invoice becomes a Tally entryThe photo is read, checked against your records, turned into a draft entry, approved by a person on your team, and only then posted to Tally.sent for approvalA photo of the invoicearrives on WhatsAppThe photo is readsupplier, items, GST, totalChecked against yourrecordsrates, duplicates, totalsA draft entry ispreparednothing posted yetA person approves itone tap on the phoneEntered in Tallyvoucher posted

In words

  1. A supplier's bill arrives as a photo on WhatsApp, or as a PDF on email.
  2. The system reads the text off the photo: supplier, items, quantities, GST and the total.
  3. It checks that against your records: a known supplier, the expected rates, not a duplicate, totals that add up.
  4. It prepares a draft entry. It posts nothing yet.
  5. A person on your team sees the draft next to the photo and approves it with one tap, or corrects it first.
  6. The system posts the voucher to Tally. Tally exports travel the other way: the system reads them every night so the reports stay current.
For your technical teamStorage, analysis, document reading, dashboards, backups
Storage
Postgres, either a managed Supabase project pinned to an Indian region or a self-hosted instance on the client's server or on-premise machine. One schema per client, row-level access by role.
Analysis
DuckDB over nightly extracts for month-end aggregation, forecasting features and ad hoc queries. Results are written back to Postgres tables that the reports read.
Document search
pgvector inside the same Postgres for price lists, policies and manuals, chunked with source and page metadata. No separate vector database to run.
Document reading
OCR and layout parsing for scanned invoices and PDFs, then model-based field extraction with per-field confidence. Low-confidence fields route to a human approval queue. Nothing posts to Tally without a person's approval.
Tally and Excel
Tally's XML interface over its local HTTP port for reading and posting vouchers, or scheduled exports where the machine is offline. Spreadsheets are read by mapped column headers with a schema check on every run.
Dashboards
Metabase, self-hosted, or Looker Studio on a read-only replica. Owners see the same tables the nightly reports read.
Backups and retention
Nightly encrypted dumps kept off-site in an Indian region. Retention is agreed per client, typically 30 daily and 12 monthly copies. Restores are rehearsed every quarter. Data is deleted at the end of a project unless the client asks us to keep it.

See what your numbers could tell you

Cleaning, forecasts and reports that build themselves, explained on one page. Or tell us which report takes the longest to prepare.

Make your numbers work

hello@pakshitechnologies.com

Monday to Saturday, 9:30 am to 6:30 pm IST