Pricing Operations
Nightly supplier pricing, reference-catalog refresh, dual-feed separation, non-fatal cache refresh, and export APIs that avoid expensive interactive rebuilds.
Work / AI-agent case study
Deep dive - AI-assisted production workHow I use AI agents without giving up production control.
The agent was not a replacement for ownership. I used it as a pair engineer for inspection, patching, debugging, validation, and repeatable operational steps while keeping scope, risk judgment, and final accountability human-led.
4Production systems showcased
Read-onlyERP audit access model
51,560Partner-pricing export rows
0Identifier collisions
The operations work touched real pricing, supplier data, scheduled jobs, dashboards, reports, and production state. AI-assisted tooling helped accelerate the mechanical parts: creating inspection commands, generating guarded patchers, checking syntax, tracing failures, comparing expected outputs, and turning operational requirements into repeatable steps.
The important boundary was ownership. I chose scope, judged blast radius, reviewed risky changes, required dry runs or read-only checks, and decided what shipped. That is the point of an AI engineering assistant in production: it should make careful work faster, not make consequential work invisible.
Project portfolio
01
02
03
04
AI agents helped turn operational requirements into production systems.
Inventory Operations
Supplier inventory ingest, dashboard visibility, and file generation where preview and export share the same canonical backend processor.
Partner Pricing
Committed partner-pricing calculations, item search, visible catalog-gap exceptions, and collision-safe CSV/XLSX exports.
ERP Inventory Audit
Restricted connector orchestration over approved reports only, with Python performing validation, joins, shortage analysis, spreadsheet generation, JSON summary, and run-status reporting.
How I use AI agents
The habits that made the work safe.
01
Inspect reality before changing it
Actual code, schemas, files, crons, permissions, service output, and production data came before assumptions.
02
Patch away from live state
Changes were staged under temp paths, anchors were asserted, syntax was checked, and backups were created before promotion.
03
Treat pricing as dangerous
Cost columns, feed identity, repricing output, and marketplace exports received extra checks because a confident mistake can scale quickly.
04
Verify with known cases
A green command was not enough. Reference items, row counts, hashes, preview/export parity, API health, and generated workbooks had to match expected behavior.
The takeaway
AI agents made careful production work faster. They did not remove the need for operational judgment.
The result is a portfolio of systems that are not just automated, but inspectable: controlled access, deterministic processing, reviewer-ready outputs, non-fatal integrations, explicit rollback paths, and human review where the business risk is high.
Let's talk
Want me to walk you through the AI-assisted workflow?
I can go deeper on how I scope agent-assisted production changes, verify outputs, and keep a human review loop where it matters.