We are addressing the critical shortage of real-life conversational datasets for voice AI.
Next-generation systems like Sesame will increasingly rely on spontaneous, natural dialogue, yet almost no large-scale datasets of this kind exist today.

Through interviews with leading companies such as Meta, PlayAI, Fish Audio, and Sony, we confirmed this urgent need.
The most surprising insight: what’s missing is everyday conversation. Despite being the most common human activity, it is largely absent from the internet. With just a smartphone, this data can be captured at scale—especially if paired with proper financial incentives. Until now, such large-scale incentive mechanisms were impractical, but crypto has made this possible.

We have developed a smartphone app where users can simply record daily conversations with friends or family and get rewarded. This creates a new collaboration space between AI and crypto.

In addition, we are collaborating with the SSSD (Scalable Spontaneous Speech Dataset) project by Meta and Carnegie Mellon University. SSSD has already collected over 700 hours of high-quality English conversations and proved the feasibility of spontaneous dialogue datasets. Together, we aim to accelerate this effort and expand it to low-resource languages, turning these datasets into a global public good with immense social value.

Currently, our app is available exclusively to early applicants, with plans to expand to the English-speaking market. While our on-chain traction is still limited, we are focused on solving a real and pressing societal problem, and we are committed to building both the infrastructure and the community to make it possible.

One of the biggest challenges was balancing the complexity of blockchain × AI operations happening in the background. On-chain transactions and AI voice analysis often run asynchronously, which risked overwhelming users. To address this, we implemented clear loading messages and alerts in the app, ensuring that the experience stayed intuitive.

We also faced issues with our native app’s OTA update feature. Occasionally, it created inconsistent states in the Privy configuration, preventing users from logging in. Since this happened mostly in production (and not during development), it was hard to diagnose. We solved it by adding error handling that clears any stale states in Privy, making the login process reliable.

Another challenge was optimizing our anti-fraud review process for USDC rewards. A fully manual review would have been costly, so we designed a system where AI performs pre-screening. This helps verify that the audio was genuinely recorded through the device’s microphone, rather than replayed from another source.

From the infrastructure side, enabling gasless transactions with smart wallets was tricky. Initially, our setup using Pimlico (Bundler) and Coinbase (Paymaster) worked on Sepolia but failed on Mainnet. After several iterations, we resolved it by implementing both components through Coinbase, which stabilized the flow.

Finally, on the frontend, it was surprisingly difficult to calibrate and visualize live voice audio levels in a way that was both accurate and user-friendly. We tuned the decibel thresholds and waveform rendering so users could clearly understand their recording quality.

0xB445f1c6FD6BD19Fe196A5894D12F46b66e5Ad29