.NET automation hub

The problem

A US-based ticket-resale business ran on fragmented systems — SkyBox, Lysted, Ticket Evolution, banking APIs, email and spreadsheets. Operations spent hours every day matching sales to inventory, recording payments, watching balances and reading presale emails: repetitive, error-prone work.

What I built

One .NET console application, dispatched by command-line argument from a scheduler, containing ~25 independent sub-programs. Every job shares the same infrastructure — configuration, Serilog logging, Entity Framework data access and a common data-access layer — so adding a job means writing the job, not another service.

• Google Sheets as the operations UI, so non-technical staff can inspect and correct data where they already work.
• Slack as the notification channel, with Block Kit summaries after each run.
• Multimodal LLM email classification (a fine-tuned GPT-4o-mini, interchangeable with Gemini) — see the email-triage case study.
• Idempotent reconciliation and bank sweeps, every transaction audit-logged.

Decisions that mattered

• No API? Use another channel. Where a vendor had no API, the data came from IMAP, parsed HTML templates (HtmlAgilityPack) or Slack conversation history.
• Explicit human-in-the-loop boundaries. When the data is uncertain the job leaves the field blank for a person, instead of guessing.
• Defensive parsing of LLM output with balanced-brace extraction and custom JsonConverters.

Outcome

Settlement reconciliation, cash sweeps and email triage run on schedule with a full audit trail across systems.

Lessons

A console app plus a scheduler beats microservices for internal automation. Google Sheets is the cheapest operations interface there is. Clear lines between what the machine decides and what a person decides prevent most downstream errors.