Independent work

I conceived, designed and built TrackBooks after chartered accountant friends described the effort of entering bills into Tally by hand. From AI invoice extraction it grew into a full accounting platform, covering sales and purchases, bank reconciliation, GST, inventory, fixed assets and reporting, and is now used by practising chartered accountants.

TrackBooks features ARIA. ARIA is its Accounting, Reporting and Insights AI agent, and carries out accounting tasks from natural-language instructions and uploaded documents. Users can describe a purchase, record a payment or upload a bill, and ARIA creates the accounting entries with validation and human review built into the workflow. It also answers questions about the books in English, Hindi or Gujarati with tables, charts and reports. Business owners with basic accounting knowledge can manage their books through conversation.

I took TrackBooks from product research and user experience through architecture, implementation and operations, with two chartered accountants advising on accounting standards. I designed it as a multi-tenant platform with a double-entry accounting engine at its core and AI agents working through controlled workflows. The architecture separates AI interpretation from financial validation and posting, with tenant isolation, audit trails and human review built in. I developed it using AI coding tools and my own workflows for testing, performance and security review.

The group had grown through acquisitions, and each subsidiary maintained its own chart of accounts. Every quarter, the finance team spent two to three weeks collecting data from subsidiary accountants, mapping their accounts to the group's reporting structure, applying adjustments and preparing financial statements for the quarterly results. The work involved repeatedly chasing submissions and reconciling different account structures, and files and revisions had to be organised so the team could track what had been received and retain the history of each reporting period.

I built a financial reporting and consolidation platform with three connected components: AI agents that map subsidiary accounts to the group's reporting structure, support accounting adjustments and prepare subsidiary and consolidated financial statements, with accountant review; a central file vault that organises source data and working files by company and reporting period, retaining revisions and historical records; and Tally integration that connects the agents to the underlying accounting data, allowing them to retrieve trial balances, ledgers and vouchers for statement preparation and investigation. Together, these provide a shared workflow from collecting subsidiary data to preparing and reviewing the group's financial statements.

A non-destructive assessment of a production logistics platform, covering the external attack surface and the code: OWASP-based web and API testing with Burp Suite, attack-surface discovery and vulnerability scanning with httpx and Nuclei, infrastructure review on AWS, and manual and AI-assisted source review with source-to-sink validation of each attack path across Java and Spring, REST APIs, SQL Server and PostgreSQL. Authentication, authorisation, tenant isolation and auditability were each tested in turn.

Confirmed four critical and two high-severity issues, including unauthenticated APIs, SQL injection and broken access control, and delivered a prioritised remediation plan covering default-deny authorisation, parameterised data access, secrets rotation and least privilege.

Users chat with or speak to an assistant named Sankhya in eight languages. All numerological calculations are done in code and the language model explains the results, which keeps answers consistent across the four model providers behind the product.

The product covers accounts and profiles, conversation, live voice calls with an animated avatar, payments through PayU and Google Play Billing, and an Android app. It has 1,000 users. I also produced the launch videos and advertisements.