Ten systems, each running today.
Industries only, no company names or figures. Every one of these started as a problem in our own group or for a client, and runs with a person approving the entries.
Equipment rental company: delivery challans and vendor bills that read themselves
Situation. Hundreds of delivery challans and vendor bills a month, each typed into the register and Zoho Books by hand, with errors found weeks later during reconciliation.
What we built. A vision model trained on the company's own challan format reads each photo, checks every field against the rent register (customer, each device, purpose, quantity) and raises anything it is unsure of to a person. Vendor bills from WhatsApp are read, matched to purchase orders, GST-checked and posted to Zoho Books only after a tap on "approve". Bank statements are parsed and classified; a Telegram bot proposes the payment entry and books it on "Book". Sales are reconciled across Tally, Zoho Books and the GST return by a repeatable method.
Result. Entry work moved from typing to approving. Every change is kept in an append-only log, so the audit trail is automatic.
Hospitality group: expense receipts from WhatsApp to the books, in three languages
Situation. Directors' expense receipts and villa purchase bills arrived in WhatsApp groups as photos, PDFs and spreadsheets, some in Kannada and Telugu, and were entered late or not at all.
What we built. A bot reads each receipt, drafts the expense entry, asks for approval in a separate group and posts it to Zoho Books with the receipt attached. Regional-language bills go through OCR and translation, are tagged to the revenue stream and property, and wait in an approval screen.
Result. Receipts are booked the day they are sent, with the image attached.
Subscription app: monthly revenue workings from payment gateways and app stores
Situation. Revenue came through two payment gateways and an app store, each with its own settlement file, fees and timing; monthly GST by state and revenue deferral were worked out by hand.
What we built. A pipeline that matches settlements to bank credits by reference, computes state-wise GST, defers revenue over the subscription period, reconciles to the bank and produces a drill-down dashboard plus an offline copy for the auditors.
Result. The monthly working is rebuilt in minutes from the source files and ties to the bank.
Car dealership: warranty claims dashboard
Situation. Warranty claims lived in spreadsheets; management could not see the approval funnel or what was being lost.
What we built. A dashboard showing the claim funnel, approval rates, rejected value and the upside from fixing the weakest step, republished automatically whenever the source file changes.
Result. One screen replaced the monthly spreadsheet review.
Coaching and consulting firm: hundreds of hours of consultant calls, analysed
Situation. Long Hindi and English sales calls were recorded but never reviewed.
What we built. Speech models tuned for Indian business calls transcribe every call; text models then rank consultants, benchmark customers, build a searchable call library and surface product opportunities and coaching points.
Result. Management can see who sells how, from the calls themselves.
Chartered-accountant practice: company incorporation documents collected by a bot
Situation. Collecting KYC and incorporation documents from several directors by e-mail took weeks of reminders and re-checks.
What we built. A bot that e-mails each director, reads the documents it gets back, checks formats, recency and consistency across documents, chases what is missing, and raises judgement calls to the operator on Telegram, with a dashboard and a management workbook.
Result. The practice sees every incorporation's status without opening a mailbox.
Food company on Odoo with four GST entities: finance operations on autopilot
Situation. Four entities, purchase bills from many sources, and a small team doing the same sweeps every day.
What we built. Morning and evening sweeps, an OCR watcher for incoming bills, banking and purchase workflows, journal and compliance checks and reporting, with every write to Odoo waiting in an approval queue and a Telegram bridge for the team.
Result. The daily finance routine runs on a schedule; people approve instead of type.
Ask your accounts: plain-English questions over live accounting data
Situation. Finance questions from founders waited for someone to build a report.
What we built. A product where the user asks in English or Hinglish, the model writes a read-only query, the database computes the exact figure, and the answer comes with a chart, the table and the query. Audit tests over the whole population and a compliance rule catalogue run on top. Refreshes hourly from Zoho Books.
Result. Live on a company's real books; the known-answer test set passes.
Private AI cluster: every project above runs on in-house models
Situation. Client financial data must not leave the office; cloud AI was not an option.
What we built. A gateway on the company's own Macs that queues requests by priority across vision, text, data-query and speech models, with a live load dashboard, heat protection, nightly security scans and a rule that no data goes to a cloud model without approval.
Result. Fifteen-plus internal applications share one set of models; no client document has left the office.
Client payroll-tax working rebuilt from source files
Situation. A client's monthly TDS working took days and its checks were manual.
What we built. An app that rebuilds the working from the source exports, applies section and rate checks, flags exceptions with codes and produces an Excel workbook built on formulas that checks its own totals. Shipped as a Windows utility with a walkthrough.
Result. The working is rebuilt in minutes and self-checks.
Which of these looks like your Monday morning?
Tell us the task and the system, and we send a scope and a price within one working day.