One and a Half Hours on a Call — Four Lines in My Notebook. That’s How Whisperer Started

On Monday, a meeting focused on architecture extended over two hours. Yet, by Thursday, someone wondered in the chat whether the discussion had ever concluded. The decision itself, which had consumed ten person-hours, seemed lost. In a notebook, there remained only four lines, including “!!” and “Ask Dima.” Meetings are often regarded as a luxury, particularly those that could have been addressed via a simple Slack message.

However, the real issue lies in the fact that conversation, arguably humanity's most expensive method of transferring information, is also one of the most ineffective ways to preserve it. The idea that sparked the inception of Whisperer revolved around the notion that transcription alone is insufficient. After all, a two-hour meeting transcript might easily amount to 18,000 words, many of which remain unread and unexamined.

The value lies not within the text itself, but in the ability to access information at a later date, such as inquiring about a specific decision made during the meeting weeks later without having to sift through pages of text. Whisperer aims to provide conversations with a memory, making the outcomes of discussions more tangible and accessible.

The most significant aspect of a meeting isn't just the summary; it's the capacity to revisit the exact moment when a decision was made. This is because summaries often oversimplify details, potentially overlooking crucial nuances, including moments of hesitation that can hold the most meaning. Whisperer can trace the evolution of a decision across multiple meetings, tracking how an initial idea of "let's try it" transforms into a fully implemented solution.

Initially, the tool struggled to identify decisions based on meeting topics, as the topic was not always present in the subject field. Instead, it resided within the parent message and the meeting graph. The tool remained silent, simply acknowledging this limitation rather than crashing. The most significant challenge in AI product development lies not in crashes, but in the silence that appears plausible.

Whisperer's translation feature emerged from the need to overcome language barriers during meetings. The tool facilitates real-time translation during live conversations, ensuring that all participants can follow along regardless of their native language. However, this requires a delicate balance, as latency can disrupt the flow of conversation.

Delays of even a single second can make the interaction feel as if it's taking place with someone on Mars. Whisperer's translation pipeline aims to minimize such latency to maintain the illusion of a natural conversation. The platform's desktop clients, designed for macOS and Windows, provide seamless integration with ongoing meetings.

These clients understand the context of the discussion, enabling the AI assistant to provide immediate assistance without the need for manual context switching. The developers recognized the importance of minimizing the human effort required to load context, as this often involves cumbersome tab-switching and clipboard management.

Instead, Whisperer's AI assistant operates within the meeting environment, offering real-time support and reducing the overall cognitive load on participants. In essence, Whisperer aims to be an integral part of the meeting experience, providing a searchable and navigable record of discussions that can be queried at any time, thereby transforming the way we approach information transfer and preservation during collaborative efforts.