assistant trigger modes
Case study 03 · Agent systems · Human-in-the-loop operations
Jarvis & Revenue Agents
Event-driven assistants that synthesise operational context, enforce guardrails and ask for approval before consequential action.
Role
Workflow architect · Builder · Operator
Status
Working automation and evolving agent framework
Automation that remains legible, governable and useful under real constraints.
01The challenge
Start with the
business problem.
A useful assistant needs to be predictable. It should know what triggers it, where work should go, what to do when something fails and which actions still need a person’s approval.
02The approach
How I
approached it.
- 01
Designed scheduled executive briefings from project, repository and task context.
- 02
Implemented deterministic South African scheduling guardrails and action-only inbox escalation.
- 03
Created a marketing pipeline with brand context, external archiving and a Discord approval gate before distribution.
03System architecture
How the parts
work together.
01
Scheduled & webhook triggers
→02
Context synthesis
→03
Schema validation
→04
Policy guardrails
→05
Approval cards
→06
Controlled action
04Evidence
Results and evidence.
executive briefing cadence
unapproved publishing paths
Figures distinguish reported business outcomes from repository-evidenced architecture. Prototype and staged capabilities are labelled to avoid overstating production status.
Selected technology
Node.jsDiscordGoogle APIsGitHubn8nDockerWebhooks
Continue the conversation