pulse

Inspiration

We wanted something smarter than Google Alerts. News apps show you headlines but nobody connects the dots across your different interests or tells you why a story matters to you specifically. This will also help in battling misinformation.

What it does

Pulse is an autonomous AI agent that tracks your interests, scans real-time sources (NewsAPI, Google News, Reddit), generates personalized intelligence briefings with "why you care" context, detects cross-interest connections, learns from your reading patterns, and lets you generate custom visualizations and magazine-style digests all through natural language.

How we built it

Next.js + TypeScript frontend with assistant-ui for chat primitives, Unkey for rate-limiting every API/LLM call, and Senso.ai as the knowledge base layer. Backend uses Vercel AI SDK with GPT-4o-mini for article analysis and cross-interest detection. Real-time updates via SSE. Four-tab UI: Command Center, Briefing, Discover, and Visualize.

Challenges we ran into

The AI SDK and assistant-ui had breaking API changes from what we expected inputSchema instead of parameters, async convertToModelMessages, no pre-built Thread component. Senso's API wasn't fully documented so we built a graceful in-memory fallback. Getting the agent to not hang on network timeouts required a force-reset safety mechanism.

Accomplishments that we're proud of

The cross-interest connection detection genuinely finds non-obvious links between topics. The Visualize tab lets you describe metrics in plain English and get real charts derived from collected articles.