Ask Forum Discord · LLM · Natural Language Search · AI Transparency

An AI-powered assistant built for KQED’s Forum, the station’s flagship call-in radio program, helping users discover and engage with three decades of archived episodes through natural conversation.

Ask Forum lives inside KQED’s Discord community, where it offers something beyond traditional search: ask a question in your own words — “When was Rachel Maddow on?”, “Are there any shows about gender?”, “What was today’s show?” — and the assistant finds relevant episodes across the archive. For select interviews with available transcripts, Ask Forum goes deeper, letting users ask questions about the content of a specific conversation. By grounding responses in the transcript itself, the system minimizes hallucination and keeps answers tied to what was actually said on air.

The project emerged from KQED’s broader AI working group, which I helped lead — a cross-functional effort with senior editors and leadership to explore where AI could genuinely serve public media audiences and where it does more harm than good. Ask Forum became the flagship experiment of that effort. A core principle: every AI interaction is transparent. If the system modifies a search query to find a match, it tells the user. When AI is involved in generating or displaying content, it’s disclosed.

Ask Forum is in active development, shaped by ongoing feedback from the Discord community it serves.

Read the full write-up on Medium →