Inspiration
Every morning, someone posts online that they need exactly what you sell — a plumber tonight, a tool that fixes the webhook delay eating their week — and by the time a dashboard or a stale intent database surfaces it, the moment is gone. I run buyer-intent infrastructure that catches those moments, and I kept thinking: nobody opens a dashboard at 7am. They talk while making coffee. The briefing had to become a voice.
What it does
Signal Brief for Alexa+ is a simulated Alexa+ experience. Pick what Alexa watches, save the profile, and ask for the brief. A deterministic analyst scores every signal PURSUE / NURTURE / MONITOR with reasons, explains what each thread means for your business, reads back the asker's own words, drafts the outreach reply, and recommends the next step — speaking a 60-word summary while the full analysis stays on screen. Ask anything naturally ("should I pursue number 1, why?"), with conversation memory, per-card deep-dives, and one-tap push to Slack.
How I built it
New repo created in-window: Express + static simulator UI, a self-hosted MCP 2025-11-25 proxy (alexa_signal_brief, alexa_next_action) in front of my Sygnal signal engine (used as an authorized third-party API), a deterministic analyst engine (verdicts, spoken-vs-screen split), and a voice loop — Fish Audio human TTS with browser fallback, mic via browser recognition with typed fallback, plus a timestamped debug trace so every mic/transcript/intent/TTS step is diagnosable. Human voice chain: Fish → Cloudflare Aura-1 → MeloTTS.
Challenges
The upstream list_signals endpoint rejected ranked queries, so I built fixtures with real source links and made mock mode the default — judges run keyless. Cloudflare's daily neuron quota died mid-testing and quota errors arrived as bare 500s the browser misreported as CORS failures, so I added CORS-safe errors, per-provider failure notes, and graceful fallbacks. Browser speech recognition mangled "brief" as "beef", so I added a mishearing map and multi-alternative checks.
What I learned
Reliability is the demo: every cloud dependency got a fallback, every failure got a plain-words message, and the product works fully offline-capable with zero keys. The spoken-vs-screen split (short voice summary, full on-screen analysis) is what makes a voice analyst feel intelligent instead of long-winded.
What's next for Signal Brief for Alexa+
Live Sygnal firehose behind the fixtures once the upstream query issue clears, an LLM re-ranker over the same signal context for freer phrasing, real Slack/CRM delivery on the push action, and multi-profile households (one Alexa, each member's own watchlist).
Built With
- cloudflare-workers
- cloudflare-workers-ai
- css
- express.js
- fish-audio
- html
- javascript
- mcp-(model-context-protocol)
- mediarecorder
- node.js
- sygnal-api
- typescript
- web-speech-api
Log in or sign up for Devpost to join the conversation.