Projects / it-support-agent
LLM triage agent for IT support
An agent that reads support emails and call summaries and opens, categorises and updates ITSM tickets, with code checks around every model step.
- Period
- 2026
- Role
- Lead GenAI and full-stack engineer
- Context
- Multi-site financial-services firm
- Cloud
- Azure
[ Architecture ]
How it works
Run a scenario to follow a request through the system. Select a component to read its role and its flows.
Problem
The IT help desk receives free-text emails and calls in several offices. Each one was read, qualified and turned into a ticket by hand, which took minutes and created duplicates.
Data
Emails, call summaries, each site’s categorisation rules and the user directory. No training data: the system relies on prompting and rules.
Approach
- Plain code handles deduplication, locks and validation; the model only understands the request, plans the actions and categorises.
- Every email is treated as untrusted, and each drafted action is checked against the source text before it runs.
- A mock mode runs the full pipeline without writing to the ITSM platform, which is how each site was validated.
Evaluation
Acceptance scenarios written with the support team were replayed on every release, and the ITSM integration was audited call by call.
Result
The agent handles email requests, and phone requests for a first office. Qualifying a request now takes seconds instead of minutes, and the support team follows and edits everything from a dashboard.
My role
Lead developer on the GenAI pipeline and the full stack, from ingestion to the dashboard and infrastructure.