Hey everyone, I've been exploring Relevance AI's platform for the past couple of weeks, specifically focusing on how it stacks up against other automation tools for quick, AI-powered workflows. I often need to build little bots that parse information and notify teams, so I decided to put it to a real test: creating a Slack bot that summarizes and alerts on new content from an RSS feed. The goal was to do this in a single, straightforward workflow.
Here's my step-by-step walkthrough of how I built it. The core idea is to have an RSS trigger, use an LLM to generate a concise summary, and then post that to a Slack channel.
**First, the components I used inside the Relevance AI studio:**
* **Trigger:** RSS Feed (configured with a tech blog's RSS URL, polling every hour).
* **AI Action:** "Generate Text" node (using GPT-4, but their other models are available).
* **Action:** Slack "Send Message" node.
**The key configuration points:**
1. **Setting up the RSS trigger** was straightforward. I just input the feed URL and set the schedule. The output is an array of items with `title`, `link`, and `contentSnippet`.
2. **The AI "Generate Text" node** is where the magic happens. I fed it the RSS item data and crafted a simple prompt. Here's the exact prompt structure I used in the node:
```plaintext
Summarize the following article into a brief, two-bullet-point update suitable for a team Slack channel.
Article Title: {{rss_feed.title}}
Article Snippet: {{rss_feed.contentSnippet}}
Provide the summary in the following format:
- **Key Takeaway:** [concise point here]
- **Why it matters:** [brief implication here]
```
3. **Connecting to Slack** required the standard Slack bot token (created via Slack's API portal). The "Send Message" node was then configured to post to my desired channel. I mapped the AI node's output as the message text, and also included the article link for easy access.
**My impressions and a couple of tips:**
* The visual builder is intuitive if you've used tools like Make or n8n. Dragging nodes and connecting them felt fluid.
* The AI node context variable (`{{ai_action.output}}`) is powerful, but remember it contains the full LLM response object. You might need to drill down to `{{ai_action.output.result}}` to get just the text string, depending on your setup.
* For more complex parsing (like extracting specific entities from articles before summarizing), you could chain multiple AI actions. I tested this by adding a preceding "Extract Information" node to pull out company names and product mentions, which worked well.
The entire build, from a blank canvas to a live bot, took me about 15 minutes. The main time was spent tweaking the prompt for the right output format and setting up the Slack app permissions. The strength here is the tight integration of the AI capabilities directly into the workflow steps without needing to call external APIs manually.
Has anyone else built similar notification systems with Relevance? I'm curious about how you've handled batching updates or adding conditional logic based on the sentiment or topic of the extracted content.
api first
api first