Clarity is a Slack bot to serve as a private communication coach, directly addressing the biggest hurdle in remote work: miscommunication. By using LLMs, Clarity analyzes your messages after you hit send, offering instant auto-edits for issues like tone and clarity.
The core technical hurdle was evaluating "good communication," which is inherently subjective and couldn't rely on standard test sets. To overcome this, we designed a sophisticated, multi-LLM evaluation pipeline. One powerful LLM synthesized the initial flagging test set, a second acted as a judge for precision and recall, and a third continuously auto-tuned the prompt for the flagger LLM, resulting in high confidence in our system's accuracy. Furthermore, we built a dedicated agent to simulate Clarity operating in various synthetic workspaces—featuring diverse industries, sectors, and personas in conflict-prone situations—to further validate its performance.
We intentionally launched Clarity as an individual-focused product to lower the barrier to entry, with the strategic goal of establishing usage before transitioning to a team-based model (M1). We are now seeking community feedback on the product before expanding our reach.
The core technical hurdle was evaluating "good communication," which is inherently subjective and couldn't rely on standard test sets. To overcome this, we designed a sophisticated, multi-LLM evaluation pipeline. One powerful LLM synthesized the initial flagging test set, a second acted as a judge for precision and recall, and a third continuously auto-tuned the prompt for the flagger LLM, resulting in high confidence in our system's accuracy. Furthermore, we built a dedicated agent to simulate Clarity operating in various synthetic workspaces—featuring diverse industries, sectors, and personas in conflict-prone situations—to further validate its performance.
We intentionally launched Clarity as an individual-focused product to lower the barrier to entry, with the strategic goal of establishing usage before transitioning to a team-based model (M1). We are now seeking community feedback on the product before expanding our reach.