Work channels
Telegram and scheduled routines
AI Operations for ShenQi
An AI system built into ShenQi’s daily work.
I built an internal system that connects company knowledge, product analytics and current work. The team receives prepared briefs, reviews the results and applies approved changes in ShenQi.
My roleArchitecture and end-to-end implementationTodayUsed by the teamConnectedKnowledge, analytics, work, Telegram and ShenQi
The task
Bring together data the team used to find across separate services.
Product performance, search data, content, email activity, documents and current work lived in different tools. The team assembled them by hand before every review. I connected the sources and set up delivery in the channels the team already uses.
Sources
Google Drive holds the main knowledge base. Analytics and work systems add current performance and active tasks.
Google DriveDocuments and company knowledgeGA4Product behaviour and performanceSearch ConsoleSearch performance and visibilityYouTubeContent performanceSendPulseEmail campaigns and activityLinearCurrent work and projects
How the work moves
Data follows a clear path from its source to a finished result in Telegram.
Orchestration
Hermes Agent, memory, routing and recurring tasks
Sources
Documents, analytics, search, content, email and Linear
ShenQi connection
Defined access to products, courses, content and pages
Product
Approved changes are applied in ShenQi
Recurring work
Workflows the team uses every day and every week.
Each morning the system gathers current performance, active work and notable product changes. The finished brief arrives in Telegram.
The system shows what changed during the week, where results diverged and which questions need the team’s decision.
Company knowledge and source data support drafts, analysis and working materials.
The team receives a prepared change, reviews it and applies it to ShenQi after approval.
Working with the product
Each workflow gets only the access it needs.
I did not give the agents unrestricted access to the application. Each workflow defines which data it can read and which change it may prepare.
The system gathers permitted context and prepares findings independently.
A result is presented before anything changes in the product.
Consequential changes run after the team has reviewed them.
Completed actions are retained, so the team can check exactly what happened later.
What I built
I connected the data sources, working channels and ShenQi product.
Technology
Hermes Agent · MCP · Rust · Laravel/PHP · Google APIs · Linear · SendPulse · Telegram Bot API · PostgreSQL · Redis
- Configured Hermes Agent, schedules, memory and Telegram delivery.
- Connected Google Drive, GA4, Search Console, YouTube, SendPulse and Linear.
- Built a dedicated integration layer between the agent and the ShenQi backend.
- Defined access to products, courses, content, pages and other ShenQi resources.
- Added change approval, safe retries and execution history.
- Established a model for tasks, revisions, review and agent handoffs.
Does your team still collect data for an important workflow by hand?
We can first map how it works today, then build one focused workflow and test its value without rebuilding the rest of the product.