HS.
All projectsEngineering case study

DIGITAL ALPHA · AI WORKFLOWS

AI accounting, ready for review

Versioned client procedures, evidence-linked reconciliation review, and a recoverable approval/release foundation.

Himanshu SharmaCase study updated

The engineering problem

An AI-generated workbook is only one part of accounting work. Reviewers also need the source documents, findings, requested changes, and a clear path to human sign-off.

My contributions

I built versioned client procedures and reconciliation review across Django and React, linking source documents, generated workbooks, evidence-backed findings, change requests, and sign-off. I integrated OneDrive source selection and asynchronous runtime callbacks with recoverable task states, durable workbook access, renewable scoped credentials, and authorization/lifecycle regression coverage.

Designing for uncertain outcomes

I implemented an approval/release foundation using a durable outbox, ordered execution steps, stable idempotency keys, and receipt-based recovery. Remote calls stay outside database transactions. I tested duplicate delivery, timeout after a successful operation, and partial completion using fixture connectors. This establishes the recovery behavior of the workflow foundation.

Keeping scope and versions explicit

My work also covered membership-derived firm/client context, client-scoped document ingestion, and versioned amendments to AI procedures with reviewer diffs, role-based approval, stale-version conflict handling, and atomic updates. Stack: Django, React, TypeScript, PostgreSQL, Celery, AWS, asynchronous runtime integrations. More professional work
All projects