When hiring platform Apna struggled to process the surge in job applications, it began experimenting with voice AI tools. But when existing solutions failed to handle the scale and complexity of India’s hiring ecosystem, the company built its own, leading to the creation of Blue Machines.
The startup came into the limelight in January after staging a one-hour unscripted live debate with Arnab Goswami on Republic TV, positioned as a “man vs machine” stress test.
“The journey to Blue Machines was a lot of serendipity in some ways,” says Abhishek Ranjan, CTO of Blue Machines, in an exclusive interaction with AIM.
Apna, one of India’s fastest unicorns, operates a job platform where millions of applications are processed. But reaching out to candidates at scale proved difficult.
Ranjan explains that at Apna, they faced a challenge: their platform receives many users and processes millions of job applications, but not everyone can be reached due to limited routing bandwidth.
At first, the team did not plan to build a voice AI platform themselves and instead experimented with existing tools from companies in India and the United States. But those systems struggled to operate reliably at Apna’s scale.
The company started with a very small team.“When we built the first version, there were about five people,” Ranjan says. Today, the engineering team has grown to around 20 people, while the overall team size stands at roughly 40 employees.
Blue Machines has not raised separate funding. As it operates under Apna Group, the company has access to existing capital. “Apna is a cash-rich company. We have not had the need to raise money,” the CTO says.
Apna became a unicorn in September 2021 after raising $100 million in a Series C round led by Tiger Global and Insight Partners, valuing the company at $1.1 billion. It had raised $193 million in total by then.
The Problem Blue Machines Solves
Apna encountered several limitations while testing existing solutions.
Ranjan stated that the first challenge was scale, noting that Apna needed to run tens of thousands of automated calls in a short time. He mentioned that many of these systems began breaking at their scale.
Another challenge was India’s real-world calling environment, where candidates answering these calls were often in a noisy environment.
And the final issue involved conversational patterns unique to India. Ranjan said that people tend to say things like ‘okay’ even if they don’t want the other person to stop, but many systems interpret that as a signal to stop the conversation.
These limitations eventually pushed the team toward building its own platform.
Tech Stack
Blue Machines is not a foundation model company. Unlike voice AI startups such as Gnani.ai and Sarvam AI, it acts as an orchestration layer for multiple models. Ranjan said the company builds enterprise workflows and infrastructure on top of these models.
“Sarvam is a foundation model company,” Ranjan says. “For us, they are partners. We sit on top of these models.” The platform adds additional context, guardrails, and workflow automation that businesses need when deploying voice AI systems.
On the compute side, the company uses Google Cloud as its primary cloud services provider. Blue Machines also has an in-house server. “It’s a streaming infrastructure built specifically for audio,” Ranjan says.
According to him, the system can handle high volumes of real-time conversations without latency or disruptions.
“It’s built in such a solid way that even if tomorrow you run a live India–Pakistan match, the architecture would still support it,” Ranjan says.
Ranjan says Blue Machines has partnered with labs like OpenAI, ElevenLabs, Cartesia and others. “You can plug in any model, and our orchestrator decides which one is best to use at a given point in time,” he adds.
With the platform expanding across industries, the company also began building its own smaller models for specific sectors, which can better understand accents and regional variations across India.
“This year, we will also put out some open source contributions, both on the model side and infrastructure side,” Ranjan concludes.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.