AI-voice agent development -- 2
Budget: $250 – $750 USD
We are looking for an experienced developer to build and train a voice-enabled AI that can answer incoming customer-support in-built App calls, understand the caller’s request, speak back naturally, and create or update a the database without a human agent. The entire AI must be trained solely based on our client's company policies, procedures, terms and conditions and business practices.
We are flexible on the tech stack—Google AI Studio, an OpenAI GPT + Whisper pipeline, or any comparable solution such as AI Agent builder is fine as long as you can demonstrate lower running costs and long-term reliability. Latency, predictable usage pricing, and ease of future maintenance matter more to me than the specific brand. For exapmple, we expect the developer (AI builder) to train the AI agent to use the most cost effective methods to answer calls or use text to speech voice/reader AND not to always call for example GEMINI or chatgpt for each function, instead the AI Agent should be trained for example to play a precorded audio file for specific responses to ensure savings instead of reading long texts. The AI Agent builder must automatically record these audio files. Therefore, the applicant we are looking for must be very experienced in prompting and giving technical prompts.
NOTE: Our client has already started using AI Google Studio vertext and they have partially built their AI but the AI Agent is not always following instructions and there are many important fine-tunings that must be conducted. Importantly, the AI Agent should be trained to self teach itself based on users ongoing behaviour and questions. For example, currently, our clients partially completed AI Agent also makes a log of all questions whcih the AI doesnt know the answer. Our client then updates this file and then uploads it to their AI's uploader files for ongoing learning and improvement.
Deliverables
• A fully trained conversational model (and prompts) tailored to my knowledge base
• End-to-end phone integration with speech recognition and text-to-speech ready for production
• Secure hand-off to our ticketing API whenever confidence is low or the caller asks for a human
• Source code, model/config files, and a concise maintenance guide so my team can retrain or tweak intents later
Acceptance criteria: during a live test the agent must handle three common scenarios in under five seconds per response, with no hallucinations and a smooth fallback when it is unsure.
If you have shipped similar phone bots, especially on ChatGPT or AI Studio, a short demo link will help me decide quickly.
This is a simple project yet important and required urgently. We understand even non developres can build such AI Agents as has been the case with our client but we want the AI to be as professional and natural as possible while keeping the costs down.
We are flexible on the tech stack—Google AI Studio, an OpenAI GPT + Whisper pipeline, or any comparable solution such as AI Agent builder is fine as long as you can demonstrate lower running costs and long-term reliability. Latency, predictable usage pricing, and ease of future maintenance matter more to me than the specific brand. For exapmple, we expect the developer (AI builder) to train the AI agent to use the most cost effective methods to answer calls or use text to speech voice/reader AND not to always call for example GEMINI or chatgpt for each function, instead the AI Agent should be trained for example to play a precorded audio file for specific responses to ensure savings instead of reading long texts. The AI Agent builder must automatically record these audio files. Therefore, the applicant we are looking for must be very experienced in prompting and giving technical prompts.
NOTE: Our client has already started using AI Google Studio vertext and they have partially built their AI but the AI Agent is not always following instructions and there are many important fine-tunings that must be conducted. Importantly, the AI Agent should be trained to self teach itself based on users ongoing behaviour and questions. For example, currently, our clients partially completed AI Agent also makes a log of all questions whcih the AI doesnt know the answer. Our client then updates this file and then uploads it to their AI's uploader files for ongoing learning and improvement.
Deliverables
• A fully trained conversational model (and prompts) tailored to my knowledge base
• End-to-end phone integration with speech recognition and text-to-speech ready for production
• Secure hand-off to our ticketing API whenever confidence is low or the caller asks for a human
• Source code, model/config files, and a concise maintenance guide so my team can retrain or tweak intents later
Acceptance criteria: during a live test the agent must handle three common scenarios in under five seconds per response, with no hallucinations and a smooth fallback when it is unsure.
If you have shipped similar phone bots, especially on ChatGPT or AI Studio, a short demo link will help me decide quickly.
This is a simple project yet important and required urgently. We understand even non developres can build such AI Agents as has been the case with our client but we want the AI to be as professional and natural as possible while keeping the costs down.