Interview with Alex Levin, the Co-Founder and CEO of Regal, part of a series of discussions.
In the realm of voice AI, Regal has made significant strides, starting its journey before the generative AI boom. The company, inspired by the power of voice for building trust with customers, has been dedicated to improving model performance over time.
Overcoming Technical Challenges
The key technical challenges Regal faced when training their voice AI agents to match or exceed human performance in natural conversations were numerous. Minimizing latency in Retrieval-Augmented Generation (RAG) was a primary concern, as even a few seconds of dead air would degrade the call experience significantly. To address this, Regal had to both speed up the retrieval process and enable the AI agents to keep talking to the customer if retrieval took longer, thus avoiding awkward pauses.
Another challenge was maintaining natural, uninterrupted dialogue flow, even when backend data retrieval was delayed. Regal tackled this issue by developing AI agents that could continue the conversation seamlessly, ensuring a smooth user experience.
Creating high-fidelity voice replicas modeled after real human voices, including investors and professional voice actors, was also a significant hurdle. Technically, this was feasible with just 5–10 minutes of high-quality audio, but it also required careful ethical considerations to ensure full consent, prevent misuse of voice cloning, and implement safety measures like safe words to distinguish real humans from AI clones.
Fine-tuning voice parameters, such as speed, tone, volume, and responsiveness, was another crucial step to optimize the user experience and ensure that the AI agents performed at or above human levels in metrics like customer containment and cost efficiency.
Applications and Future Developments
Regal provides AI voice agents for sales, support, scheduling, and collections. However, the future looks promising as these agents are expected to evolve to become indistinguishable from humans and exceed human agent abilities in the next year.
Reinforcement learning, while not currently used due to legal concerns, may be introduced in 13 months. Consent is required from the person whose voice is being cloned, and bad actors who clone voices without consent should be shut down.
Regal's AI advisors, created to pay it forward to founders, have been used for advice on product roadmaps, pricing models, and more. Alex Levin, the founder of Regal, was previously a leader at Angi and Handy.
The Regal platform features a no-code builder, real-time analytics, A/B testing, and built-in compliance for regulated industries, making it an attractive solution for businesses. Companies that lean into AI agents are expected to drop their costs and improve customer experience faster than anticipated.
The AI advisors have firm-specific advice and philosophies tied to the individual VCs, as their prolific writings were uploaded into the respective AI Agent Knowledge Bases. This unique feature sets Regal apart, offering businesses not only advanced AI technology but also valuable insights from some of the industry's leading minds.
[1] Source: Regal's Whitepaper 2021 [3] Source: Regal's Blog Post - "The Future of AI in Sales: Regal's Predictions" 2022
- To address technical challenges, Regal improved their artificial-intelligence (AI) agents by minimizing latency in Retrieval-Augmented Generation (RAG), enabling seamless conversation during delayed backend data retrieval, and creating high-fidelity voice replicas that include careful ethical considerations.
- In the near future, Regal's AI voice agents are anticipated to surpass human agents in performance, and the introduction of reinforcement learning, with proper consent, may be a significant development in 13 months.
