AI Agent Development Services

Description

Our operations team runs a nightly routine that nobody enjoys: pull yesterday's orders, cross-check them against three supplier portals, flag mismatches, and email a summary before 8am. It takes one person about ninety minutes and it is the reason we cannot hire in that role without also hiring a backup. We want an agent that does the pull, the cross-check and the draft, and stops at the point where judgement starts. It should hand a human the mismatches with the evidence attached rather than resolving them itself. If it cannot reach a supplier portal it should say so in the summary instead of silently reporting fewer mismatches. Two of the supplier portals have no API. Scraping them is in scope; pretending they have an API is not.

• Build the nightly agent run as a scheduled job with retries and a dead-letter path
• Integrate the one supplier API and scrape the two portals that have none, with breakage alerts
• Produce the 8am summary email, with mismatches grouped and evidence linked
• Make unreachable-source failures explicit in the output rather than absent from it
• Add a small admin view so ops can re-run a night without a developer

• Scheduled agent in our environment with monitoring and alerting
• Summary email template, reviewed with the ops lead
• Admin re-run view
• Runbook covering portal layout changes and credential rotation