The context
Field consultants hit sales anomalies they cannot resolve alone: a rejected contract, inconsistent data, a procedural question. With no dedicated tool, those requests travel through informal channels and get lost.
U-Hub structures that flow: a ticket, a skill, a named responder, and a knowledge base designed to surface the answer before a ticket is even created.
What I built
Structured ticketing
types, categories, tags, statuses and predefined replies for recurring cases.
Skill-based routing
assignment to groups of agents able to handle the subject, rather than a single queue.
Realtime discussion
chat rooms attached to tickets, with attachments.
Full-text search
a dedicated search engine over tickets and the knowledge base.
FAQ and documentation
an article base with tags and sub-tags.
Agent rating
scoring of responses, to make support quality measurable.
Notifications
realtime alerts and mobile push.
Architecture & technical decisions
Search was the structuring subject. Classic relational search becomes unusable as soon as ticket volume grows and you want typo tolerance. A dedicated search engine was integrated with continuous indexing from the database — the goal being a response fast enough for an agent to use live during a call.
Skill-based routing models agents as capability groups rather than one queue. That is what stops a highly specialised ticket from sitting behind twenty trivial ones.
Like the platform's other services, the application runs on a persistent application server and broadcasts its events in realtime, so chat and status changes appear without a reload.
The service stays deliberately autonomous: it exposes an API other applications call to open a ticket, so support never becomes a blocking dependency of sales.
Technical challenges
Keeping the index in sync. An external search engine introduces a second source of truth: incremental indexing plus a full rebuild path when drift occurs.
Realtime with several people. Multiple agents can view the same ticket simultaneously — updates must converge without silent overwrites.
Measuring without degrading. Agent rating had to stay an improvement tool, which means aggregating over time rather than exposing raw individual scores.
Attachment volume. Screenshots and documents stored as objects, with controlled access.
Meaningful canned replies. Structuring recurring cases to cut handling time without turning support into impersonal automated answers.