InSuporte: Support Dashboard and AI Insights
My role
I built the app end to end, from the Convex data model to the AI insights, as part of the AI Enablement team.
InBot's support team tracked tickets across separate dashboards, fed by scripts tied to one person's account. InSuporte brings everything into a single app: day-to-day operations and AI-driven trend analysis.
Operations dashboard: total, resolved, open, urgent, average and worst-case resolution time, filterable by company, type and assignee.
ClickUp mirror: incremental sync while someone is looking, plus a nightly full reconciliation.
Daily rollups: metrics never scan the ticket table; every change updates that day's aggregate.
AI insights: conclusion, evidence, action and chart for each period. Every text starts as a draft and is only published after a person reads it.
Frozen numbers: what went out by email is preserved, and discrepancies show up as warnings, never as silent rewrites.
Internal InBot system: the code is proprietary and the screenshots use fictitious data.
Before
2 dashboards and 4 automations: an Apps Script API with anonymous access, reports sent from one person's account, and a monthly close that once published the year-to-date total as if it were the month.
The problem
Support numbers were scattered across two dashboards and four automations: a static front end, an Apps Script API published with anonymous access, and reports triggered from one person's account. The insights panel was public, and one monthly report once published the year-to-date total as if it were the month.
With 1,800+ tickets and growing volume, leadership needed numbers it could trust and a clear read on trends.
The solution
I built a single Convex app that mirrors the ClickUp ticket list as read-only. Company and type are parsed from the title using rules ported from the old scripts and covered by tests. Metrics come from daily rollups updated on every change, without scanning the ticket table.
Period insights (biweekly and monthly) are generated by AI in the same format as the skill the team already used, but they start as drafts: a person reviews and publishes them, and the numbers behind each release are frozen. Access now requires company login.
Decisions and trade-offs
A read-only mirror of ClickUp instead of a new source of data. Metrics computed from daily aggregates, without scanning the table. AI insights start as drafts and are only published after a person reviews them, with the numbers of that release frozen.
The results
InSuporte replaced 2 dashboards and 4 automations, removed the dependency on one person's account and closed anonymous access to the data. Leadership receives human-reviewed insights, each with evidence and a recommended action.
Results by the numbers
- 1.800+Tickets mirrored
- 2Dashboards replaced
- 4Automations replaced
- 100%Analyses reviewed by a person