Blue Machines AI launches Aurora for Indian BFSI calls

Blue Machines AI has launched Aurora, a multilingual speech-to-text model for the banking, financial services and insurance sector, aimed at real-time financial conversations in India.

Aurora is designed to handle language switching, code-mixed English and regional language speech, as well as telephone audio with background noise. It is also built to recognise sector-specific terms and entities, including EMIs, policy numbers, transaction IDs, monetary amounts, interest rates and account references.

According to Blue Machines AI, internal benchmarking on BFSI sample datasets found a Semantic Word Error Rate of 1.51% for English, 2.43% for Hindi BFSI conversations and 5.52% across multilingual speech. The company also reported a BFSI Entity Error Rate of 4.23% for data points such as policy numbers, payment dates and transaction references.

The tests covered banking, lending, insurance, collections and customer servicing conversations. The audio included Indian English, Hindi, Hinglish and multilingual speech, along with regional pronunciation patterns, telephony distortion and background noise.

Sector focus

The launch reflects a broader push by technology suppliers to tailor speech recognition tools to specific industries rather than rely on general-purpose transcription systems. In financial services, errors in capturing repayment terms, identity details or transaction values can affect customer service, compliance checks and follow-up actions.

Blue Machines AI said Aurora was evaluated against other speech-to-text models using the same audio inputs and scoring method, and reported stronger accuracy with lower latency in streaming BFSI use cases.

The company also disclosed processing speed and scale figures. In internal tests, Aurora recorded 236 milliseconds P50 latency and supported up to 960 concurrent real-time streams per H100 at a 320 millisecond operating point, rising to 2,400 concurrent streams per H100 at a 1.12-second operating point.

Nirmit Parikh, Founder and CEO of Blue Machines AI, said the model was built around the realities of Indian financial interactions.

"Aurora reflects our commitment to building sovereign AI infrastructure for Indian enterprises.

"India's financial conversations do not happen in a single language or follow a standard script. When AI misunderstands an EMI amount, policy number or repayment commitment, it can change the customer outcome. By building Aurora in India, we are giving financial institutions speech intelligence designed for how their customers naturally communicate, while ensuring greater control over their data, models and customer interactions," said Parikh.

Custom training

Blue Machines AI said it has developed training pipelines that allow Aurora to be adapted with customer-authorised enterprise data. The aim is to help institutions teach the model their own product names, terminology, geographic references, accents and interaction patterns.

Internal evaluations showed that customer-specific retraining could cut recognition errors by 40% to 45% on institution-specific datasets compared with the base model, according to the company. The approach reflects broader demand among banks and insurers for AI systems that can be tuned to internal processes rather than rely solely on standard models trained on broader public data.

Abhishek Ranjan, Chief Technology Officer of Blue Machines AI, described the technical approach behind the system.

"Building speech intelligence for BFSI requires more than generic transcription," said Ranjan.

"Aurora's cache-aware FastConformer encoder and streaming transducer decoder enable it to retain context while processing speech incrementally. The model has been optimized for multilingual and code-mixed speech, low-latency inference and high-concurrency environments. Crucially, it is evaluated on its ability to accurately recognize the entities that drive financial workflows, not merely the surrounding sentences. In internal throughput tests, Aurora supported 960 concurrent real-time streams per H100 at a 320 ms operating point and 2,400 concurrent streams per H100 at a 1.12-second operating point."

Wider platform

Aurora sits within Blue Machines AI's broader customer experience software for enterprises. It can be used across customer acquisition, onboarding, lending, collections, servicing, insurance, claims and support workflows, the company said.

It can also be deployed on a managed cloud, inside an enterprise virtual private cloud or on-premises, according to Blue Machines AI. The range is intended to address data residency, governance and security requirements that remain central for regulated financial institutions handling sensitive customer records and transaction data.

Blue Machines AI is part of the Apna Group and focuses on AI tools for customer interactions across voice, chat, WhatsApp, web and email. Aurora adds a speech recognition product built around the linguistic complexity of Indian financial conversations and the operational demands of high-volume contact environments.