Kolkata-based Mihup is a vernacular voice interface platform founded by Tapan Barman, Biplab Chakraborty and Sandipan Chattopadhyay in 2016.
“The technology came to India a little late, and the global brands find it challenging to deal with the linguistic diversity of the country. Our conversational AI platform integrates Hindi, Hinglish and a mix of other regional languages like Tamil, Bengali, Telugu etc. We want to change the way Indians and the world at large interact with machines and make technology accessible to all,” said Tapan Barman, Co-founder & CEO, Mihup.
Mihup is an abbreviation of ‘May I Help You Please’. The startup has partnered with Tata Motors to bring multi-language voice-controlled agents in cars. The product, which is currently available in Tata Nexon and Altroz XZ (O), supports a mix of vernacular languages and can be run even without internet connectivity.
What’s the USP?
“Mihup has created three products which are built on our proprietary platforms of Speech to Text; Natural Language Processing; Dialogue Manager; and Text to Speech. Our Voice AI products cater mainly to the contact centre and automotive space, with the potential to be incorporated onto any platform,” said Tapan.
- VIA (Virtual Interaction Analyst) – To analyse all inbound/outbound customer interactions to generate actionable business insights.
- AVA – Call Center (Automated Virtual Agent) – To facilitate automated and accurate management of customer queries through a human-like AI interface.
- AVA – Auto – a multi-language vehicle AI assistant which offers various features such as car control, media and entertainment.
Most conversational AI platforms are based predominantly on US English. “However, what we are doing is fundamentally altering the conversations by integrating Hindi, Hinglish and other mixes of regional languages into the conversation,” said Tapan. Moreover, the product doesn’t necessarily need an internet connection at all times and can seamlessly work in low-connectivity areas and even offline.
Tech behind
“Mihup’s technology has a low carbon footprint. It provides Voice AI on the edge. It can work reliably on less compute than the industry standard (10- 20x lesser), and Voice AI on edge ensures zero connectivity requirements,” added Tapan.
- Automatic Speech Recognition (ASR) – neural network-based models that have been optimised to achieve the state of the art accuracy and can run on a modern single-core CPU.
- Query understanding framework – Uses various modules for NLU activities such as machine learning techniques involving statistical approach-based linear regression, boosting techniques.
- Information Retrieval – uses custom data structure using hierarchical clustering-based hashing technique to perform complex queries on large volumes of data with filters involving Metaphone, fuzzy matching, context-based matching and give output in real-time.
- Dialogue Manager powered by Text to speech (TTS) – uses neural network-based models for natural language generation (NLG)
Tech stack
- C++, Java and Android — For products that require edge inference
- NodeJS, Python and Golang for backend
- Angular for frontend
- Postgres for database
- Redis for caching
- RabbitMQ for message queuing
Hiring at Mihup
“Our people are our strength, and we are committed to create a workplace of trust and respect. We are always looking for people with diverse experiences, the capability to think differently and to put in a collaborative effort towards the success of Mihup. We have plans to expand our team size to 100+ with cross-functional hires, an almost 70% increase from the existing team of 62. We recently also conducted interviews in some management and engineering colleges like IIMs, IITs and NITs,” said Tapan.
Future plans
Recently, Mihup raised $1.5 million in Series A round led by Accel Partners, Ideaspring Capital, leading investor Rajesh Jain’s firm Core91 VC and Jayant Kadambi, Founder & CEO, YuMe Networks. Plans are underway to raise Series B funding by mid of FY 21-22.
“Mihup has already analysed over 100 million customer interactions and offers 100% interaction analysis to give a precise understanding of the “Voice of the Customer.” Our business has grown 60X, and we have added customers from various domains – from Fortune 500 companies to technology unicorns. We have witnessed a 5x jump in our client base in a matter of just one year. We have also onboarded ten new clients in the last quarter,” said Tapan.
Mihup is competing with both national and international companies, including Uniphore, Observe AI and Callminer.
“We have plans to expand into the international market with the next round of fundraising. Our goal is to touch $3 million ARR in FY 21-22. We are also adding new capabilities to our interaction analysis product VIA, which will make it a complete go-to Voice AI solution for improving the efficiency of call centre interactions and agents. Features like real-time assistance, auto-scoring are a part of the product roadmap and would be available in FY 21-22,” concluded Tapan.