AI content automation · n8n
Pegregator
A multi-stage pipeline that turns curated Telegram channels into topic-based audio digests: dedicated stages for collection, AI classification and processing.
Current status: Founder project. Workflows archived; presented as an architecture and delivery case, not as a live production system.
Problem
Following many Telegram channels by hand is slow and noisy: duplicates across sources, irrelevant posts and no way to consume the useful part on the go.
What was built
- A progressive web app that collects posts from curated Telegram channels across 12 topics.
- A multi-stage n8n pipeline with a claim → classify → finish pattern, staging, retries and debugging logic.
- Gemini-based classification plus deduplication and clustering in n8n, followed by text-to-speech audio digests.
- Modular workflow design so each stage can be tested and replaced independently.
What it demonstrates
- Turning an informal information habit into a product with a defined pipeline
- Reliability thinking: staging, retries and failure isolation
- Product thinking combined with hands-on implementation
My role
- Product decisions, scope and prioritization
- Tool and vendor choices
- Acceptance, incident handling and cost/risk control
- Hands-on workflow architecture in n8n
Stack / methods
Evidence boundary
Much of the implementation and debugging was AI-assisted. No user, revenue or uptime figures are claimed here.