A focused look at the systems I built with AI-assisted tooling: pricing, inventory, partner exports, ERP audit workflows, and the safety habits that make production automation dependable.
Before
Supplier files, reference catalog data, and marketplace costs could drift without an easy review path.
After
A nightly Python/API pipeline ingests, validates, prices, caches exports, and flags risky findings before they ship.
Cached export packages start downloads in seconds
Before
Supplier stock truth lived across uploaded files, templates, exports, and dashboard views that could drift apart.
After
One backend processor owns preview and export, with cached overview data and supplier-specific file generation.
Supplier output verified at 16,221 workbook rows
Before
Partner pricing needed item eligibility, marketplace shipping data, committed calculations, and exception visibility.
After
A committed pricing module exposes search, rules, exports, and a separate remediation list for catalog gaps.
51,560-row main feed with zero identifier collisions
Before
Inventory exceptions were spread across item, balance, sales order, purchase order, and movement data.
After
An AI agent orchestrates a read-only ERP reporting workflow while Python produces deterministic spreadsheet and JSON outputs.
Guarded connector access, no record writes
I use AI agents as pair engineers for inspection commands, guarded patchers, syntax checks, structured debugging, and repeatable deployment steps. The production rule stays simple: inspect reality, stage away from live files, validate with known cases, back up, promote, and verify.
End-to-end systems built with Python, JavaScript, APIs, scheduled jobs, file processing, validation, reporting, monitored infrastructure, and AI-agent orchestration. Select a workflow to see how its components work together.
A working slice of the kind of module I build. Edit a cost, price, or stock level and the margins, risk flags, and totals recalculate instantly - the same deterministic rules a reviewer would rely on. Everything here is synthetic sample data; no client information is shown.
The nightly Pricing Update Pipeline - how it fails safe on cost columns and ships the smallest safe change set.
Read the case study -> InventoryThe Inventory Visibility System - day-over-day change detection at scale, with a sample morning change report.
Read the case study -> AI Agent WorkflowThe operating model behind the production work: inspect, stage, validate, back up, promote, and verify.
Read the case study ->I'm happy to talk through the design decisions behind any system here. Book a time or send a note.