Dasha

The company teaching machines to talk on the phone

Most of the AI world spent the last few years racing to build things that type back at you. Dasha placed a different bet: that the interface that matters is your voice, and the place it matters most is a phone call.

Dasha is a voice AI platform. Stripped of jargon, that means it gives software developers the tools to build agents that can pick up a phone, understand a caller, and respond in a voice close enough to human that people often forget they are not talking to one. The agents can qualify a sales lead, screen a support request, confirm an appointment, or run a survey - the repetitive, script-adjacent conversations that fill call center shifts.

The company was founded in 2018 by Vladislav Chernyshov and Ilya Stupakov, engineers whose roots trace back to Novosibirsk, Russia, and who eventually planted a headquarters in New York. From the start they made an unusual decision: rather than stitch together off-the-shelf speech recognition and text-to-speech, they built the core stack in-house. The reason was a single number - latency.

That obsession explains a lot about Dasha. In a text chat, a half-second delay is invisible. On a phone call, it is the difference between a conversation and an interrogation. Humans expect the other side to jump in almost instantly, to hear when they have finished a thought, sometimes to talk over each other. Dasha's engineering is aimed squarely at that gap: real-time voice-activity detection, in-house speech recognition, and a text-to-speech engine that - notably - was taught to add "ums," "ahs," and even breathing sounds. It sounds like a flaw. It is the point. Real people are not fluent, and an AI that never hesitates is an AI that sounds like a robot.

What people can actually do with it

Dasha is aimed at developers and the businesses they build for. A team can use it to automate outbound sales calls, staff an inbound support line that never sleeps, screen and qualify leads before a human ever gets involved, book appointments, or run post-sale satisfaction surveys. Documented use cases stretch across banking and finance, healthcare, insurance, and retail - including a dental practice, a bank's self-service line, and even an in-game voice assistant for a console title.

The way you build these agents is where Dasha diverges from the crowd. Instead of a drag-and-drop flowchart, the company created a programming language. DashaScript is an event-driven, declarative language written for one purpose: describing how a machine should talk on the phone. Around it sits Dasha Studio, a Visual Studio Code extension for building and debugging agents, and a Node.js SDK that wires all of it into a developer's own systems. The pitch is that voice AI should be infrastructure you build on, not a product you are locked into.

Where it sits in a crowded market

When Dasha launched in 2019, its competition was named Google Dialogflow, Amazon Lex, Microsoft LUIS, and Google Duplex - the demo that made headlines by booking a haircut over the phone. TechCrunch covered Dasha's arrival, and in 2020 Gartner named it a Cool Vendor in Conversational AI Platforms, an early nod before most people had ever spoken to an AI on a call.

The landscape has since filled in. A new wave of voice-agent platforms - Vapi, Retell AI, Bland AI, Synthflow, PolyAI, ElevenLabs, Cognigy - now competes for the same developers. Dasha's answer is partly its head start and partly its architecture. The platform is LLM-agnostic: bring GPT, Claude, or Gemini, and Dasha handles the voice. The wager is that the language model is becoming a commodity while the real, defensible work lives in the voice layer - the recognition, the synthesis, the sub-second timing that makes a call feel alive.

How it makes money is straightforward for the category: business-to-business software, sold as usage-based access to run calls at scale through the cloud API, plus enterprise deployments. Publicly, Dasha raised a $2 million seed in 2019 led by RTP Global, with FinSight and MTS StartUp Hub also participating. A later $2.5 million Series A has been reported but does not appear in public funding databases, so it is best treated as unconfirmed. The team sits at roughly 30 people, research-heavy, with a lineage of PhDs and competitive-programming alumni that shows in the decision to build so much from scratch.

Chernyshov has never been shy about the stakes. He has said conversational AI "will replace people everywhere technology allows" - a line that is either bracing or unsettling depending on which end of the phone you are on. What is clear is that Dasha planted its flag on voice years before the rest of the industry decided voice was worth the trouble. Now that the rest has arrived, the company that spent those years grinding on latency and disfluencies is finding out whether being early was the same as being right.