With proprietary speech stacks tuned for real-time enterprise use, the startup is gaining BFSI traction by prioritising speed, trust, and natural conversations over model size.
Challenging the Big Model Bias in Voice AI
As generative AI races forward, Smallest.ai is quietly carving out a lane of its own—by going small. Co-founded in 2023 by Akshat Mandloi and Sudarshan Kamath, the San Francisco-headquartered startup is building compact, proprietary speech models designed specifically for real-time enterprise voice use cases, not content generation.
- Unlike ElevenLabs or Cartesia, Smallest.ai focuses on speed, security, and realism, not just scale.
- Its small language models (SLMs) support text-to-speech, speech-to-text, and speech-to-speech pipelines.
- The models are tuned to understand emotional nuance, breathing patterns, and native languages—vital for BFSI and regulated sectors.
“For real-time conversations, even milliseconds matter. Large models can’t always keep up,” said Mandloi.
Built for Enterprise, Not Just Demos
Most voice AI tools today are trained for content generation—think audiobooks or marketing videos. But live business conversations need a different stack: low-latency, high-security, and platform flexibility.
Smallest.ai’s edge lies in re-architecting layers for performance, training from scratch rather than relying on open-source off-the-shelf models. Its third-gen TTS engine supports English, Spanish, and top Indian and European languages. A full-stack approach includes:
- Memory and intelligence layers to contextualise conversations
- Cloud-agnostic deployment, with optional on-premise hosting for privacy-sensitive clients
- Real-world use cases in customer support, banking voicebots, and AI advisors
Mandloi believes deep control over architecture is key:
“We don’t just fine-tune—we fundamentally rethink the layers to prioritise speed and conversational flow.”
BFSI: The Perfect First Vertical
While speech interfaces are booming globally, India’s $1T digital economy by 2030 makes it a massive playground. In sectors like BFSI, where users prefer native-language voice over typing, latency and trust are everything.
- Smallest.ai now serves multiple listed banks and fintechs in India and the US (names undisclosed).
- Focus is on narrow, high-frequency voice use cases, with gradual scope expansion.
“Texting is not a natural modality for most users in BFSI. Voice is,” Mandloi said.
Adoption may be slower in regulated industries, but trust compounds once security and performance are proven.
Infrastructure Agility + Security by Design
Enterprises remain cautious with AI, not due to efficacy doubts but due to concerns around deployment flexibility and data protection.
- Smallest.ai’s solution can run on AWS, Azure, or customer-owned infrastructure.
- Security was embedded from Day One—not retrofitted.
- Deployment often begins with pilot use cases, building confidence over time.
“You can run it on your cloud or ours. That flexibility builds trust,” Mandloi said.
The inflection point came in mid-2024, when a LinkedIn demo of Smallest.ai’s voice model went viral. Since then:
- Revenue has grown 300% YoY in the US and 150% in India.
- It raised $8 Mn in seed funding from Sierra Ventures, with earlier backing from 3one4 Capital, Upsparks, and DeVC.
- The model is monetised via annual enterprise licenses and usage-based SaaS subscriptions.
In a crowded voice AI market, Smallest.ai isn’t trying to be everything—it’s trying to be fast, efficient, and enterprise-friendly.
The Smallest.ai Philosophy: Build Fast, Stay Focused
While competitors raise massive rounds, Mandloi believes distribution now matters more than model building. “Get to customers fast. Ship often. Stay lean,” he advised fellow founders.
Still, hiring remains a challenge—speech tech talent is rare, and the team remains small and selective.
“It’s a needle-in-a-haystack problem. But once we find the right people, we know it fast,” he said.
Mandloi is also candid about India’s funding landscape:
“Risk capital for deeptech is still limited in India. We were lucky with timing and traction.”
TL;DR:
Smallest.ai is building small, high-performance speech models tailored for low-latency enterprise voice use. With traction in BFSI, proprietary tech, and 300% US revenue growth, the startup is proving that in voice AI, speed and trust may matter more than size.
AI summary:
- Smallest.ai builds compact, proprietary voice models for real-time enterprise use
- Founders: Akshat Mandloi and Sudarshan Kamath
- Core use case: BFSI voice AI (native language, low latency, secure)
- Raised $8 Mn in seed funding (Sierra Ventures, 3one4 Capital, others)
- Supports multilingual speech-to-text and speech-to-speech pipelines
- Revenue up 300% in US, 150% in India YoY
- Infra-agnostic; offers on-premise deployment for regulated clients
