URGENT : Python Expert for LiveKit & Azure Voice AI for Final Deployment & Bug Fix)

URGENT: Senior LiveKit & Azure Voice AI Expert Needed (Final Deployment & Bug Fix) Job Title: Final Deployment & Bug Fix: LiveKit Voice Agent + Azure OpenAI + Azure Speech (Python) Project Type: Short-Term / Troubleshooting / Launch Support Tech Stack: LiveKit (Python SDK), Azure OpenAI (gpt-5-mini), Azure Speech Services (STT/TTS), FastAPI, Supabase. - THE PROJECT OVERVIEW We have built a massive, fully-containerized Enterprise Voice AI & Outbound Dialer system for 500+ manufacturing factories. The system connects via standard SIP trunks, routes calls using FastAPI & Supabase, and runs isolated LiveKit workers for both Inbound Reception and Outbound Sales. The architecture is fully built, cleanly written, and 100% dockerized. We recently migrated our LLM/Speech stack from Gemini/Sarvam to Microsoft Azure (Azure STT + Azure OpenAI + Azure TTS) to achieve sub-700ms latency and eliminate JSON serialization errors. The issue: The SIP calls are connecting perfectly to the LiveKit cloud/self-hosted server, but there is dead silence. We are hitting specific Azure configuration/integration errors on the backend. If you are a LiveKit expert, this is a very fast, straightforward task. The code logic is sound; we just need an expert to fix the plumbing and get the agents talking. - THE CURRENT INBOUND ERRORS (Why it's silent) You will be fixing these specific errors inside our [login to view URL] worker: Error 1: Azure OpenAI Configuration (404) APIStatusError: Error code: 404 - {'error': {'type': 'invalid_request_error', 'code': 'DeploymentNotFound', 'message': 'The requested deployment name does not exist...'}} (Likely a simple mismatch in our .env variables regarding the Azure OpenAI Deployment Name vs. Model Name). Error 2: Azure TTS Configuration (400) failed to synthesize speech: message='Bad Request', status_code=400, retryable=False (Likely an issue with the region formatting, parameter mapping, or the specific Indian Neural Voice (en-IN-NeerjaNeural) eligibility in our Azure Speech resource). Error 3: LiveKit Pass-Through Drop dropping pass-through signal — no stream available (A direct consequence of the LLM/TTS failing to generate an audio stream). - YOUR EXACT DELIVERABLES 1. Fix the Inbound Agent & Tune Latency: Debug the Azure API connections in livekit_agent.py. Ensure Azure STT, Azure OpenAI (gpt-4o-mini), and Azure TTS (en-IN-NeerjaNeural) stream perfectly together. Tune the Silero VAD (Voice Activity Detection) parameters to achieve our target < 700ms conversational latency. 2. Deploy & Test the Outbound Pipeline: Our outbound architecture is fully written but untested. You will need to verify the flow: Ensure the Async Pacer ([login to view URL]) successfully fires the CreateSIPParticipantRequest to LiveKit. Ensure the Outbound Worker ([login to view URL]) picks up the room, speaks first when the user answers, holds the conversation, and successfully triggers the log_call_disposition function tool at the end of the call. - WHAT YOU WILL BE GIVEN You will not have to guess how the system works. You will be provided with: The complete, clean Python codebase ([login to view URL], [login to view URL], [login to view URL], [login to view URL], [login to view URL]). The [login to view URL] and Dockerfile. Access to our Azure portal on Anydesk/Zoom screen-share. Access to our LiveKit Cloud / Self-hosted server through Anydesk/zoom screen-share. A SIP phone number to test inbound/outbound dialing directly from your cell phone. - THE IDEAL CANDIDATE You have actively deployed LiveKit Voice Agents to production. You know the livekit-plugins-azure and livekit-plugins-openai packages inside and out. You understand how LiveKit handles SIP trunking ([login to view URL]). You are highly proficient in Python asyncio. No agencies. We are looking for a sharp, solo engineer who can look at these logs, spot the issue, and push the fix today. To Apply: Please start your proposal with the word "NEERJA" so I know you read the errors. Briefly explain what you suspect is causing the 400 Bad Request on Azure TTS or the 404 DeploymentNotFound on Azure OpenAI, and your estimated time to fix this and test the outbound pipeline.

Project ID: 40570437