AI IT asset platform

Refresh planning starts with a client's device inventory: a spreadsheet or CMDB export that can run past 20,000 devices. Before anyone can plan a refresh or price it, each model needs its specs and lifecycle dates. Done by hand, that research took around two days per inventory.

As lead developer, I built a platform that turns those inventories into specs, refresh plans and ROI reports. It reached production in about two months and is used by around fifteen consultants. Research that took two days now takes about an hour.

Underneath is a LangGraph pipeline in five stages. Inventories are ingested and normalised, model names are mapped to canonical products with a consultant confirming the ambiguous ones, and research draws on vendor APIs with Bing grounding for the gaps. A verification gate sits before anything ships as a refresh plan or ROI report, with telemetry in App Insights.

Click through the stages below to see what each one does.

The rule that shaped the design: unverified fields come back as TBC instead of a confident guess. A refresh plan built on a made-up end-of-life date is worse than one with a visible gap, because nobody knows to check it. About 80% of dates resolve automatically; the rest are marked TBC for a consultant, not filled in.

Production meant more than the pipeline. The platform also runs on:

Around $600K in savings per client, research time down from about two days to an hour, and roughly 80% of lifecycle dates resolved without a human.