Building Todoist Ramble: How Doist Turned Voice Braindumps into Real-Time Task Capture

The podcast details Duists development of Rumble, a voice-to-task AI feature for Todoist, designed to simplify unstructured task capture by allowing users to speak freely, with the system extracting actionable tasks from audio input. This innovation addresses the "cold start" challenge, where users brainstorm ideas before documenting them, by bridging the gap between unstructured ideation and structured task creation. The feature leverages large language models (LLMs) for real-time audio processing, enabling direct speech-to-task conversion without relying on traditional transcription workflows. Key design goals include reducing friction, enhancing usability, and ensuring seamless integration with Todoists existing tools, such as task editing and labeling. The development was informed by user research, competitor analysis, and insights into how people prefer pen-and-paper brainstorming over digital tools, aiming to align AI capabilities with real-world needs rather than novelty.

Technical implementation focuses on real-time processing, where audio is streamed to a backend microservice interacting with LLMs to generate and update tasks dynamically. The system avoids text-to-speech for simplicity and reliability, offering visual and auditory feedback to confirm task creation or edits. Challenges include parsing ambiguous or fragmented user input, handling date formats, and balancing creativity with precision in LLM outputs. Solutions involve prompt engineering, standardizing language inputs, and iterative testing across multiple languages and scenarios. Future directions include expanding Rumble to handle multimodal inputs like text and images, improving contextual awareness for team collaboration, and enhancing automation by integrating with external tools like Slack or Microsoft Teams. The overarching philosophy emphasizes user-centric problem-solving, prioritizing intuitive interaction and simplicity to externalize tasks effectively while accommodating diverse human workflows.