Doist, a leader in remote-first work solutions, has introduced Ramble, a feature designed to elevate the user experience of its popular task management app, Todoist. Inspired by a memorable scene from The Devil Wears Prada, Ramble enables users to seamlessly convert their stream-of-consciousness speech into organized task lists without the need for typing or formatting.
Identifying Challenges: The development of Ramble involved overcoming several technical hurdles, including:
- Fast and accurate real-time communication with tool-calling capabilities.
- Multilingual support that accommodates slang and accents.
- Non-deterministic output testing and semantic validation.
- Reliable audio handling across various browsers.
Implementing the Solution: Ramble was built on the Gemini Enterprise Agent Platform, utilizing its advanced capabilities for real-time interactions. The feature processes raw audio directly, allowing Gemini to manage language detection and speech recognition efficiently. This approach minimizes latency and enables autonomous task management as users speak.
Technical Architecture: The back-end structure of Ramble is designed for scalability and flexibility, featuring:
- A provider-agnostic streaming layer.
- A dictation module for one-way audio.
- A brain dump module for capturing unstructured thoughts.
- A conversation module for future bi-directional audio features.
Outcomes and Impact: Ramble has significantly improved task management for users, delivering clear and concise task breakdowns. The collaboration with Google has facilitated a deeper partnership, enhancing support and sustainability for high API usage. The success of Ramble has encouraged Doist to further explore AI applications across productivity workflows.
Future Directions: Doist plans to expand Ramble’s capabilities beyond task creation, considering features that enhance the entire productivity journey, from planning to automation.
Testing and Validation: Rigorous testing involved both structural and semantic validation to ensure that the model accurately understood user intent. This process included real-world recordings in multiple languages to assess performance and maintain quality across different linguistic contexts.
Ultimately, Ramble represents a significant advancement in how users can manage their tasks, joining Todoist’s existing features to offer an unparalleled task capture experience.
