AI Voice Assistant Secures Funding for Music Production Startup

Following the successful refactoring, our specialist proceeded with further development, building out a suite of core features, including:

• AI utilization screen, which is the primary screen for users to issue commands to the Voice Assistant
• Settings screen, which allows users to customize the application according to their needs and audio devices, as well as personalization of the Voice Assistant
• Audio device configuration, allowing musicians to configure their input and output devices
• Picture-in-picture mode, with seamless transitions and an always-on-top feature

A notable custom solution we provided was the manual adjustment and configuration of sound card input and output channels for audio recording, since Web Real-Time Communication (WebRTC) used in the app doesn’t support this accurately. Ultimately, this solution established a correct and automated input/output configuration.

We enhanced the app by implementing a multi-agent AI system. Initially, all user tasks were handled by the OpenAI GPT-4o-mini-realtime-preview model, which proved to be costly. To optimize operational expenses, we leveraged Agent as a Tool and FunctionTool features, distributing tasks across various AI models. For example, a more powerful model like GPT-4o-mini or Google Gemini processes a user’s voice command and translates it into a concise text prompt.

This prompt is then passed to a more cost-effective LLM that can execute the specific textual task efficiently and affordably. This approach creates a collaborative squad of AI models, each playing a specific role to execute user commands effectively while significantly reducing overall operational costs up to 46%.

Our DevOps specialist took charge of the application’s end-to-end infrastructure deployment, setting up CI/CD pipelines, orchestrating architecture builds, managing Apple notarization, and facilitating beta testing for the minimum viable product (MVP).

Addressing the time-consuming and costly nature of Apple notarization, our specialist came up with a fast and cost-effective solution. Instead of performing the entire notarization process on macOS, we executed builds on macOS and acquired notarization signatures via a Python script on LinuxOS. This approach significantly reduced the client’s financial overhead.

By adopting a unique trunk-based development workflow, we streamlined the DevOps process and saved significant time and costs. Instead of creating separate builds for development, staging, and production, we used a single, tested build. This was made possible by leveraging a remote configuration, which allowed us to seamlessly transition the staging build to production without rebuilding the application.