About

roject Overview We’ve built a Python-based AI voice agent using LiveKit Cloud that handles both inbound and outbound SIP calls through Twilio. Now we need to make it smarter — the agent should automatically: Navigate multi-level IVR menus (pressing DTMF digits or speaking keywords). Detect when it reaches a human and seamlessly switch to our normal AI prompt. Handle both inbound and outbound calls using one Twilio number without conflicts. Escalate to a manager or human rep when required. You’ll work directly on our existing LiveKit agent (Python) to implement this IVR navigation layer and improve reliability for production use. Responsibilities Integrate and tune LiveKit SIP Gateway with Twilio Elastic SIP Trunking. Implement IVR auto-navigation logic: Use ASR transcripts to detect IVR options. Send DTMF (RFC2833 / SIP INFO) or spoken responses automatically. Detect and confirm human pickup. Add logic for session handling (unique room per call) so one number can manage both inbound/outbound calls. Build structured prompt-to-action decision logic (LLM-driven or rule-based). Improve latency, barge-in, and VAD performance. Log every action (digits pressed, phrases spoken, transitions). Write clean, production-ready Python code with comments. Requirements ✅ Must have Strong Python background (async, WebRTC, real-time audio). Hands-on experience with LiveKit (Agents SDK / SIP Gateway). Deep understanding of Twilio SIP / Elastic SIP Trunking and PSTN call flows. Knowledge of DTMF signaling (RFC2833, SIP INFO). Familiarity with ASR/TTS pipelines (Whisper, Deepgram, Azure, etc.). Ability to debug SIP traces (Wireshark, sngrep, etc.).