Equal scales its Call Screening Assistant to enterprise volume with Sarvam

Every day, millions of phone calls reach Indians, including loan agents, delivery drivers, bank calls, insurance renewals, unknown numbers, and the calls that actually matter. They come in every language, every accent, and a mix of languages spoken across a country as vast as India. For most Indians, every one of these calls comes straight through, with no way to know what matters until they pick up. Equal AI is changing that by building India’s Personal AI category, starting with an AI Call Assistant that manages every call before it reaches you. Every call screened by Equal AI is powered by Sarvam’s Speech Recognition APIs, which are built to understand India in the languages people actually speak.

Background

Equal AI is building India's Personal AI category, starting with an AI Call Assistant that handles every incoming call on the user's behalf. When a call arrives, Equal AI answers, speaks to the caller in their language, identifies who they are and why they are calling, and delivers a full summary and recording once the call ends. Processing crores of calls every month, Equal AI is the first product in India to democratise the idea of a Personal Assistant.

The Problem

Equal AI end customers operate across India, generating voice interactions in a wide range of Indian languages and regional accents. For an AI Call Assistant to be useful in this environment, the underlying transcription has to perform consistently across that linguistic variety, handle code-switching between English and Indian languages, and remain reliable under high-volume, real-time conditions.

The priority was to establish a speech-to-text foundation that could:

• Transcribe voice interactions accurately across multiple Indian languages and regional accents.
• Handle code-switching between English and Indian languages reliably.
• Maintain accuracy in noisy, real-world audio environments where call quality varies.
• Operate at enterprise scale with consistent latency and uptime under load.
• Scale elastically as new enterprise clients onboard to Equal's platform.

Why Sarvam

Production usage demonstrates a strong product fit. Equal AI has progressively upgraded through Sarvam's Speech Recognition Model model versions as quality has improved, and now runs Saaras V3, Sarvam’s speech recognition model as the primary model behind the majority of its transcription workload.

As Equal AI's user base scaled across the country, the volume of voice interactions flowing through the platform grew rapidly and so did the linguistic diversity. Every new user brought new accents, new dialects, new mixes of Indian languages in a single call. At Equal AI's scale, every call had to be transcribed, understood, categorised, and responded to within seconds, in the caller's language, without fail. That is where the partnership with Sarvam began.

The Solution

Equal AI's Call Screening Assistant runs entirely on Sarvam's Speech to Text stack. Across production, the workload is distributed across the full breadth of Sarvam's Speech Recognition Model capabilities, including standard transcription, real-time streaming, bulk batch processing, and speaker diarization, with Saaras V3 as the primary model powering the majority of calls.

Transliteration is the dominant mode in production, accounting for roughly 97% of Equal AI's Speech Recognition Model calls. Spoken Indian-language audio is converted into Roman script, which downstream systems in the financial workflow can consume directly. Transcription, translation, code-mix, and verbatim modes are also used where the use case requires them, giving Equal AI the flexibility to match the appropriate Speech Recognition Model configuration to each interaction.

Equal AI has progressively adopted successive Sarvam Speech Recognition Model releases as they have shipped, moving through model versions in lockstep with the quality improvements Sarvam has rolled out. That trajectory reflects a partnership where Equal AI's scale on the platform has grown alongside the maturity of Sarvam's Speech Recognition Model models.

The Impact

• Scale: Approximately 34 million minutes of audio processed every month.
• Growth: Production volume on Sarvam’s Speech Recognition models grew roughly 420x between October 2025 and May 2026, from about 555,000 calls in October to 234 million calls in May. By June 2026, this had increased further to a run rate of approximately 285 million calls per month, reflecting sustained enterprise confidence in Sarvam for Indian-language speech recognition at scale.
• Month-on-month trajectory: Following the production scale-up in October 2025, usage grew at triple-digit rates through early 2026, with 113% growth in November, 264% in December, 176% in January, 608% in February and 240% in March. Growth then moderated to 38% in April and 14% in May as usage reached a sustained high-volume run rate.
• Reliability: 99.98% success rate across approximately 129 million Speech Recognition Model calls in the most recent 30-day window.
• Model adoption: Saaras V3 powers approximately 85% of Equal AI’s Speech Recognition Model calls, reflecting continued adoption of newer Sarvam model versions as speech recognition quality has improved.

Looking Ahead

What the partnership has already demonstrated is that Indian-language voice AI can operate dependably at a scale that very few API customers in India have reached. As Equal AI continues to expand the reach of its Call Screening Assistant across financial workflows, and as Sarvam continues to advance its multilingual Speech Recognition Model capabilities, the partnership is well positioned to shape how voice AI is deployed across India's fintech ecosystem.