A new Show HN project called Heyalo is challenging the assumption that AI in meetings should be seen and not heard—or rather, heard but only after the fact. The developer behind the project, posting under the handle heyajoe, shared their journey of building an in-call AI assistant that's designed to surface relevant information during a meeting, not afterward when it's too late.

The Problem With Existing Tools

The insight driving this project is painfully relatable for anyone who's sat through meetings while fumbling for answers they knew existed somewhere. "We use Attio for call recordings," heyajoe explains in the original post. "And they work great... that's kind of the problem." The summary arrives with the answer to the question they couldn't answer well during the meeting, even though all that information existed the entire time—in their notes, their CRM, earlier in the same conversation.

Why Making AI Shut Up Is Harder Than Making It Smart

Here's where it gets interesting from a technical perspective. The hard part wasn't building an AI that could retrieve relevant context or synthesize answers on the fly. It was preventing that AI from becoming yet another meeting participant who talks too much, interrupts flow, and derails conversations with well-intentioned but poorly-timed contributions. Designing restraint into an LLM-powered system turns out to be a fundamentally different engineering challenge than designing capability.

Real-Time Context vs. Post-Meeting Summaries

The distinction Heyalo is chasing represents a meaningful shift in how we think about AI in professional settings. Traditional tools like Otter.ai, Fireflies.ai, and even built-in platform transcription focus on capture and retrieval—they record everything and let you search later. What heyajoe is describing is something closer to a real-time research assistant who whispers answers only when asked (or when context strongly suggests the information would be immediately useful), rather than a chatbot that wants to engage.

The Technical Challenge of Contextual Restraint

Building such a system requires solving several hard problems simultaneously: understanding what's being discussed in real-time, determining what relevant information exists in connected systems (notes, CRM, previous conversations), deciding when an answer would be genuinely helpful versus disruptive, and presenting that information without breaking conversational flow. Each of these is its own research problem; getting them to work together in a live meeting context raises the complexity considerably.

Key Takeaways

  • The real pain point isn't lack of AI tools—it's having answers AFTER you needed them DURING
  • Making an LLM talk LESS is harder than making it smarter or faster
  • Contextual restraint and timing matter more than raw capability for in-call assistants
  • Integration with existing data sources (CRM, notes, past conversations) is essential to be useful

The Bottom Line

This project hits on something the AI assistant space has largely ignored: meetings aren't documentation exercises—they're collaborative problem-solving sessions where every interruption carries a cost. Heyalo's focus on restraint over capability might just be the right bet, even if it's harder to build.