AskSafe

Inspiration

We've all been there. Sitting in a 300-person lecture, confused about something the professor just said, but too anxious to raise a hand. Or in a company all-hands where the CEO asks "any questions?" and the room goes silent — not because there are no questions, but because no one wants to be the one to ask.

This silence has real consequences. Students fall behind and don't recover until exam results make it obvious. Employees disengage because their concerns feel invisible. Community members leave townhalls feeling unheard.

We built AskSafe because we believe the best questions are the ones people are afraid to ask — and technology should make it safe to ask them.

What it does

AskSafe is a real-time anonymous Q&A platform for any one-to-many setting:

For participants:

Join with a 6-character code, verify you're human via World ID (no personal data stored)
Ask questions anonymously by typing or whispering (ElevenLabs voice-to-text)
Rate your confusion with a single tap when the host checks in
Upvote question clusters that matter to you
Get a post-session email summary of what was covered

For hosts:

See a live dashboard with animated confusion gauge, participant count, and question count
View a per-slide confusion timeline that shows exactly when the room got lost
AI clusters similar questions into topics with rich summaries — address confusion, not chaos
Four response options: AI-drafted answer (editable), share a link/resource, flag for later, or custom response
End-of-session report with AI insights, confusion spikes, resolution rate, and student feedback

For organizations:

Works for university lectures, company townhalls, board meetings, community forums, training sessions
Proof-of-human prevents bot spam and ensures 1-person-1-vote integrity
Post-session analytics help improve future sessions
AI agents available via ASI:One chat and OmegaClaw for on-the-go access

How we built it

Frontend: Next.js 16, Tailwind CSS, shadcn/ui, Recharts, Socket.IO client
Backend: Python FastAPI, Socket.IO server, Motor (async MongoDB)
AI Agents: 3 Fetch.ai agents on Agentverse with Chat Protocol — Confusion Monitor, Question Clustering, Insight Report. Same logic powers both ASI:One chat and the app backend.
AI: Google Gemini for semantic clustering, explanations, reports, email summaries, and feedback filtering
Identity: World ID IDKit v4 with server-side RP signing for proof-of-human verification
Data: MongoDB Atlas with 7+ collections and aggregation pipelines
Voice: ElevenLabs Scribe API with Web Speech API fallback
Email: Resend for post-session student summaries
Real-time: Socket.IO for live confusion updates, question counts, cluster broadcasts, upvotes
OmegaClaw: Integrated via Telegram — professors can query session analytics from their phone

Challenges we ran into

World ID v4 requires RP signatures generated server-side — built a Next.js API route using the official signing utility
IDKit WASM binary needed manual copying to the Next.js public directory
OmegaClaw Docker required WSL which crashed on Windows — had to reinstall WSL from scratch
Agentverse agents use the uAgents protocol (not REST) — integrated agent logic directly into the backend so the same intelligence serves both interfaces
Making AI clustering useful required richer prompts — moved from "short labels" to detailed summaries that capture distinct sub-topics within each cluster

Accomplishments we're proud of

A complete, working product that solves a real problem people experience every day
Three AI agents that serve dual purpose: ASI:One chat for discovery AND app backend for production
World ID Orb verification working end-to-end with real QR code scanning
OmegaClaw successfully discovering AskSafe, querying our API, and providing its own analysis of confusion patterns
The "Address Cluster" flow where AI suggests but the human decides — technology augmenting, not replacing, the host
Post-session feedback loop that gives hosts actionable insights immediately, not weeks later

What we learned

Multi-agent architectures where the same logic serves multiple interfaces (app + ASI:One + OmegaClaw)
World ID v4 RP signing flow for proof-of-human in web applications
Real-time Socket.IO patterns for live classroom-scale interaction
How to make AI clustering genuinely useful (rich summaries with context, not just keyword labels)
The importance of the "AI suggests, human decides" pattern — professors trust the tool more when they have final say

What's next for AskSafe

Deploy frontend to production with custom domain
Native mobile app for hosts to monitor sessions on the go
Integration with LMS platforms (Canvas, Blackboard, Google Classroom)
Multi-language support for international classrooms and global townhalls
Analytics dashboard for recurring sessions — track improvement over time
Enterprise tier for companies running regular all-hands and training sessions

Use Cases

University lectures — Students ask questions they'd never raise their hand for
Company all-hands — Employees surface concerns anonymously to leadership
Board meetings — Junior members contribute without hierarchy pressure
Community townhalls — Residents ask hard questions about local issues
Training sessions — Trainees flag confusion without slowing the group
Conference Q&A — Audience members submit questions during talks