A chartered accountant who builds the systems he runs — revenue recognition, controllership and reporting — in SAP, Python and AI, where they actually have to close the books.
Everything here was built inside a working finance function — against locked-down SAP, hard close deadlines, and no IT team to hand it to. That's the constraint I design for: it has to run in production, not in a sandbox.
I'm a chartered accountant who runs finance operations and builds the tools that fix them. Designing the system and building it are usually two jobs, held by two people — and the accounting logic gets lost in the handoff. I keep both in one head: diagnose the accounting, design the control, build the tool — a reconciliation engine, an automated statement builder, an AI agent, a dashboard pipeline.
The result is finance that runs like software — governed, repeatable, audit-ready. Everything below is a working rebuild on synthetic data, open to inspect.

I'm Deepak Sharma — a chartered accountant who runs finance operations and builds the tools that fix them. I don't write the code by hand: I use AI — Claude Code — to turn the accounting logic in my head into working software, fast. I've spent my career in the seat where the month actually closes: revenue recognition, controllership, order-to-cash, and the SAP configuration underneath all of it.
Somewhere along the way I stopped waiting for tools to arrive and started building them — a reconciliation engine here, an automated statement builder there, an AI agent for the work nobody enjoys. The catalogue on this page is that habit made public: every item a working rebuild on synthetic data, open to inspect.
Years inside pharmaceutical and complex biotech finance taught me the function in depth — hands-on through the detail, and leading from the front while managing large teams through it. That depth is the foundation.
I'm working toward running a finance function end to end — not just signing the numbers, but owning the system that produces them. But the building is the part I enjoy — whether it's teams, products, or process.
— Deepak Sharma, CA
Each entry is a working tool, rebuilt from scratch on synthetic data so it's fully open to read — organised by the capabilities behind them.
A local, deterministic desktop app that reconciles a payment run against history and produces a tamper-evident audit report — without ever exposing bank data to an LLM.
A deterministic two-file reconciliation app that handles any recon — vendor, customer, bank — with a header-mapping layer that absorbs mismatched files and remembers the settings.
A three-layer engine that turns a trial balance into Schedule III / IND AS statements where every figure cascades from one source of truth.
Turns a raw ledger extract into an Opening → Movement → Closing reconciliation and a cash bridge, with an AI layer that explains every material swing.
Wrote the requirements for the project-system module and found two configurations that would have baked a revenue-recognition error into the system.
Replaced one tangled ledger with a three-register model that separates invoicing, revenue and receivables — IND AS 115-compliant by design.
Reads an email thread, judges the right escalation level, and drafts the follow-up — with the accounting nuance built in.
A rule-based budget-request and approval system that auto-clears what's within limits and escalates only the exceptions.
Designed the commercial, tax, customs and ERP treatment for shipping inventory to an overseas warehouse with no sale — and no permanent establishment abroad.
Python that drives the SAP GUI directly to automate high-volume, error-prone tasks — with read-back validation and a full status log.
A pipeline that pulls structured, finance-ready data out of commercial contracts and vendor invoices — text or scanned.
An agentic OS for a finance function — drop a task, and it routes to the right specialist agent, runs the skill, pulls the resources, and produces the deliverable.
It was built and deployed the same way as everything in it — with AI, by me. The simplest working example of the rest.
I'm curious about open source and love connecting with like-minded people. If you'd like to talk, reach out below.