privacy1st.app

An AI-powered call screening platform that shields your personal phone number from spam, scams, and robocalls — fully Agentic workflow letting only the calls that matter reach you.

Comment

Privacy1st

Inspiration

Demo: https://drive.google.com/file/d/1ccF4nUBn_0tbZLsLT62Wu0B_xqsHRVUa/view?usp=sharing 4.8 billion robocalls hit Americans every month and that's only Americans, consider we Vietnamese, Chinese, Russian, friend across the Globe being disturb a lot, and we stop to use our phone number. We stopped answering our phones entirely — and started missing calls from doctors, schools, and real opportunities. Existing call screening apps "solve" this by harvesting your contacts and selling your data to brokers. We wanted a solution that actually respects privacy: no app install, no data harvesting, no carrier switch — just an AI receptionist powered by Agora's real-time audio infrastructure that answers your phone and only lets the right people through.

What it does

Privacy1st is an AI-powered call screening platform built on Agora's Conversational AI Engine. When you activate it by dialing a simple carrier forwarding code, incoming calls are routed to an Agora RTC channel where the AI receptionist handles the full conversation — screening callers, blocking spam, and letting VIPs through. The Agora CAI Engine orchestrates the entire STT → LLM → TTS pipeline with sub-400ms end-to-end latency, so conversations feel natural. You can monitor any screened call live from your browser via Agora WebRTC and take over the conversation at any time. Currently, the core feature is phone number redirect-based AI screening via carrier call forwarding. No app to install, no carrier to switch.

How we built it

  • Real-time Audio Core: Agora Conversational AI Engine for managed STT → LLM → TTS pipeline, Agora RTC SDK for browser-based live call monitoring and takeover
  • Telephony: Twilio for DID numbers, SIP trunking, and PSTN bridging into Agora channels
  • Voice AI: ElevenLabs Flash v2.5 for natural multilingual TTS (29 languages), Deepgram Nova-3 for real-time STT
  • LLM: OpenAI GPT-4o-mini for caller classification, conversation handling, and function calling
  • Frontend: Next.js 16 (App Router), React 19, Tailwind CSS 4, Agora RTC SDK (Web) for live monitoring dashboard
  • Backend: FastAPI (Python 3.12), SQLAlchemy with async PostgreSQL, Agora token generation for secure channel access
  • Security: Per-user AES-256 encryption, JWT auth, zero third-party data sharing

Challenges we ran into

  • Twilio account false suspension: Our Twilio account got immediately suspended on first use due to automated fraud detection — before we could even make a single test call. The appeal process took valuable hackathon time and was frustrating. This reinforced our architecture decision to keep Twilio as a thin PSTN bridge and let Agora handle the heavy lifting for real-time audio and AI conversation.
  • Agora SIP trunk provisioning: Connecting Twilio's PSTN layer to Agora RTC channels via SIP required careful configuration to maintain audio quality without codec conversion overhead.
  • Real-time latency tuning: Getting the full pipeline (caller → Twilio → Agora → STT → LLM → TTS → caller) under 400ms end-to-end meant optimizing every handoff point between services.
  • Social media AI streaming: We planned to extend the AI receptionist to Instagram and TikTok live using Agora rooms, but these platforms require partner API approval we haven't received yet — this remains on the roadmap.
  • Privacy-first architecture: Building per-user encryption and data isolation while keeping real-time audio performant added complexity at every layer.

Accomplishments that we're proud of

  • Zero app install required — activation is literally dialing a carrier code from your phone
  • Sub-400ms conversational AI powered entirely by Agora's real-time infrastructure
  • Full live call monitoring and takeover from a browser via Agora WebRTC — no mobile app needed
  • Per-user AES-256 encryption with zero data sharing by design
  • 29-language AI voice support out of the box
  • A complete platform (auth, dashboard, analytics, call logs, flow builder, AI agent builder, CRM) built during the hackathon We built a whole pipeline that is automated and Agentic for Agora voice AI. Apparently, it can call tools; it can call web search; it can do everything with tiny fish, with Google API, like any kind of API. It can do a lot of things, so we create a box, a pipeline that integrates everything in the center of its, is Agora. The best part is it can be use with OpenClaw too.

(over 300$ burned in the process of vibe coding lmao)

What we learned

  • Agora's Conversational AI Engine dramatically simplifies real-time voice AI — managing the STT/LLM/TTS pipeline as a service saved us weeks of custom orchestration work
  • Carrier-level call forwarding is a surprisingly powerful and underutilized product primitive
  • Twilio's fraud detection is aggressive with new accounts — plan for suspension appeals during hackathons
  • Privacy-first design is harder but builds genuine trust — users shouldn't have to trade one privacy problem for another

What's next for privacy1st.app

  • Social media AI screening: Extend the AI receptionist to Instagram DMs, TikTok, and other platforms via Agora streaming rooms (needs platform API approval so we build this demo purely on phone number)
  • Agora SIP trunk migration: Move DID provisioning closer to Agora's infrastructure to reduce Twilio dependency and lower per-call costs
  • Smart caller learning: AI that learns from your call patterns to auto-classify callers over time
  • Team/business mode: Multi-agent routing with Agora multi-channel orchestration for shared numbers
  • Open API: Let developers build custom screening logic on top of Privacy1st's Agora-powered voice AI platform

Built With

Share this project:

Updates