AWS AgentCore phone voice agent coded in Python -- 2

Summary Job Title: AWS AgentCore phone voice agent coded in Python 5 days to do *** important Provide minimum delays between question – answer – question for conversation dialogs *** important provide very good voice to text even in noisy environmet and with user interuption capabilities Use or https://aws.amazon.com/about-aws/whats-new/2026/09/Amazon-Connect-Customer-A2A/ Posted on: Sep 22, 2026 with AI agents outside Connect Customer, over the open agent-to-agent (A2A) protocol through text or bidirectional voice AWS expanded the protocol to support voice streaming Customer phone ↕ Amazon Connect Customer ↕ A2A voice streaming ↕ AWS AgentCore (Python) ├── Speech-to-text ├── Bedrock LLM └── Text-to-speech Or just idea Phone provider: Twilio can't connect directly to AgentCore, because its stream address doesn't allow the login signature AgentCore needs. Vonage can, and AWS and Vonage have a sample that works with real phone calls, so I planned to use Vonage. Voice code: the Pipecat library (version 1.12) has everything we need in Python. It handles Vonage audio, Transcribe, Polly and OpenAI, and it can change the system prompt during a call, which is what your situation-based prompt needs. OpenAI on AWS: GPT-5.6 Luna can be called through Bedrock from Canada (Central). Requests go to Canada or US regions. Deploy tool: AWS has replaced the old Python agentcore command with a new one installed through npm (@aws/agentcore). https://us-west-2.console.aws.amazon.com/bedrock-agentcore/home?region=us-west-2# Job Type: Fixed-Price Very good natural voices for LLM use low cost fast models from openai or gemini or make suggestion **** No Terraform or similar techniques, only simple browser aws platform development *** use AWS from local windows Project Overview **** Develop Voice Real Estate agent (sell homes) **** For testing provide realistic properties descriptions from zillow.com or zolo.ca or realtor.ca etc. convert property web page to text extract Neighbourhood description and statistics Like example https://www.realtor.ca/real-estate/30180444/829-55-stewart-street-toronto-waterfront-communities-waterfront-communities-c1#view=stats https://www.realtor.ca/real-estate/30180444/829-55-stewart-street-toronto-waterfront-communities-waterfront-communities-c1#view=neighbourhood https://www.realtor.ca/real-estate/30180444/829-55-stewart-street-toronto-waterfront-communities-waterfront-communities-c1 I am looking for an experienced AWS and Python developer to build a scalable Voice AI Agent architecture using AWS Bedrock AgentCore. This is fundamentally an educational project: your goal is to build a fully working example system, document it exhaustively, and teach me how to reproduce, modify, and deploy it completely independently from scratch. looking for a one tenant Voice AI platform where agents can be created and managed, with Canadian phone number, isolated skill files, call memory/logs, and admin management, built on AWS Bedrock AgentCore + Python, along with complete documentation, videos, testing, and handover. You may use existing open-source GitHub projects as a foundation, provided the final deliverable meets all my requirements and is fully documented. agent answer questions about properties and persuade user to buy and book viewings intergrade some useful booking service and send message by email and sms text to both user and realtor full log such call is transcribed and session logs System Architecture & Core Examples We need a one Canadian phone voice agent system acting as customer service caller phone number ID identification email interface for admin to provided gmail address : add or remove files for skills; send by email logs of conversations , block phone numbers for incoming call The Use-Case Examples: Think of these as sales agents for stores or services , where store has many unique products, for example : Or Case 1: Food Catering Or case 2: Cleaning Services Or case 3: Refrigerator Repair Or case 4 : Real Estate agent (sell homes) Expected Interaction Flow: At the beginning of the call, the agent asks the user what they want to talk about (essentially asking which file to add to system prompt ). Example A (Case 3): A user calls for Store 3 and says, "I have Refrigerator abc123." The agent must dynamically retrieve the specific .txt skill file for "Refrigerator abc123" and use this skill for the remainder of the session. Example B (Case 2): A user calls for Store 2 and says, "I need to clean a sofa." The agent chooses the specific .txt skill file on how to clean a sofa and uses it for the session. Example C (Case 3) : user wants to talk about specific home for sale . Find skills file with specific address what user tells you . Each .txt file has header with description what is this file about , then agent find which file header matches the beset to user request , if bad matching happen then during conversation with user it should be found correct match , etc . Usual skills router implementation Also if user wants then conversations returns to state to choose what property to discuss **** Also search of matching properties by search criteria for example all properties from your portfolio with one bedroom , test with 10 properties in portfolio *** Technical Features & Requirements Voice Integration: Phone call voice chat using a Canadian phone number (+1). One phone number serves exactly one agent/store. Strict Data Isolation: agent has its own dedicated group of .txt knowledge files. Session Memory: The agent must retain context during the call (e.g., if the caller says "My name is Peter," the agent calls him Peter for the whole session). There must be no memory between different sessions. Logging: Detailed logs must be generated and saved for every individual call/session. Technical Stack Language: Python AI Framework: AWS Bedrock AgentCore Telephony: use Amazon Connect - low cost AWS services compatible with AgentCore , I am open to discuss other options like Chime SDK, o