Why fine-tuning matters

Your AI agent discovers content based on the topics and sources you configure. Out of the box, it casts a wide net. Fine-tuning narrows that net so you get fewer but more relevant results - saving your team time and improving the signal-to-noise ratio.

Setting up topics

Topics are the primary way to guide your AI agent. Each topic is a phrase or keyword that tells the agent what to look for.

Good topic examples

• "React Server Components best practices" (specific and actionable)
• "B2B SaaS pricing strategies" (clear domain + focus)
• "Kubernetes security hardening" (technical + scoped)

Topics to avoid

• "Technology" (too broad - you'll get everything)
• "News" (no focus - results will be random)
• "Good articles" (subjective - the AI can't judge this)

Adjusting source quality

Not all sources are equal. Configure your agent to prioritize high-quality sources:

1. Go to your group's AI Agent settings
2. Review the sources your agent is pulling from
3. Add preferred sources that consistently produce relevant content
4. Exclude sources that generate noise

Tip: Start with a few trusted sources and expand gradually. It's easier to add sources than to filter out noise from too many.

Using feedback loops

The best way to improve relevance over time is to interact with the content your agent discovers:

• Like items that are highly relevant - this signals quality to the system
• Skip items that miss the mark - patterns in skipped content inform future filtering
• Add notes explaining why something is relevant - this context improves future matches

Advanced configuration

Combining topics with tags

Set up your agent to auto-tag discovered items based on which topic triggered the discovery. This lets you:

• Filter agent results by topic area
• See which topics produce the most valuable content
• Identify topics that need refinement

Scheduling

Adjust how frequently your agent runs:

• Daily: For fast-moving topics like news or trending tech
• Weekly: For evergreen topics where daily updates would be noise
• On-demand: Trigger the agent manually when you need fresh content

Measuring relevance

Track these signals to know if your agent is well-tuned:

Next steps

• Setting Up Your AI Agent if you haven't configured one yet
• Building a Full Research Pipeline for end-to-end automation
• Automation with n8n for advanced workflow triggers