Two Financial Services Companies Are Backing India's AI Research. What Do They Want to Build?
Bajaj Finance and Pennant Technologies have invested a combined $4 million in Bengaluru-based Soket AI. The partnership aims to develop indigenous foundation models and agentic AI solutions for banking and other regulated industries.
Why Are Two Financial Services Companies Investing in an Indian AI Research Startup?
For banks and financial institutions, adopting AI comes with an additional requirement: the technology must be reliable, secure and able to operate within strict data and governance controls.
Bajaj Finance and Pennant Technologies are looking to address this opportunity through a strategic investment in Bengaluru-based AI research company Soket AI.
Announced on November 18, 2026, the two organisations have invested a combined $4 million, with approximately $2 million contributed by each.
Bajaj Finance, part of Bajaj Finserv, made its investment through Finserv Intelligence, Bajaj Finserv's group-wide initiative in applied research and innovation. Pennant Technologies, an investee company of Bajaj Finance, invested the remaining amount.
Unlike a conventional venture-capital investment, this partnership also has a practical business objective: connecting Soket's foundation-model research with the financial-services and enterprise technology expertise of its investors.
The companies expect to explore AI applications in areas where data security, reliability and control over technology are particularly important.
The Research Behind Soket AI's Project EKA
Founded in 2019, Soket AI is led by Abhishek Upperwal, Founder and CEO, and Dr. Mayank Singh, Co-founder and Chief Science Officer.
Upperwal studied at the Indian Institute of Science, Bengaluru, while Singh is an Associate Professor at IIT Gandhinagar and holds a PhD from IIT Kharagpur.
The company's flagship research initiative, Project EKA, focuses on developing indigenous foundation models for mathematics, coding, reasoning and multilingual AI.
Its first model, EKA-1, was released under the IndiaAI Mission on November 3, 2026.
According to the Ministry of Electronics and Information Technology, EKA-1 has 26.3 billion parameters and was trained on approximately 7.5 trillion tokens. It supports all 22 scheduled Indian languages alongside English and is designed for mathematical reasoning, code, logic and agentic applications.
The model is built as an open, sparse hybrid language model. Soket's wider Project EKA initiative also explores AI for Indian and Global South languages and programming.
The IndiaAI Mission selected Soket to develop indigenous foundation models, providing access to national AI computing infrastructure and support for research expenses.
For the company, the objective is to develop globally competitive AI systems that can also be deployed in sensitive environments where organisations need greater control over their data and technology.
Loop: Giving AI Agents a Way to Work Through Longer Tasks
Alongside its foundation-model research, Soket has introduced Loop, an open-source agentic coding harness.
Loop is designed to give AI agents a structured environment for interacting with tools, maintaining sessions and working on software repositories. Its documentation describes a Rust-based system with a stateful agent, an interactive terminal interface and Model Context Protocol (MCP) support.
Multi-agent orchestration is available as an optional capability.
This is different from releasing another language model. While EKA-1 provides the underlying model capabilities, Loop provides an environment for agents to use models and tools to carry out tasks.
Together, the two products reflect Soket's broader direction: developing models, inference infrastructure and agent capabilities that organisations can operate within their own environments.
Where Banking and AI Research Come Together
The strategic investment becomes particularly interesting when looking at what the two partners intend to build together.
Pennant Technologies develops AI-powered lending and banking technology for financial institutions. Its collaboration with Soket will include developing and evaluating foundation models, agentic tools and inference infrastructure.
The companies plan to explore banking-specific language models and AI agents for:
KYC and onboarding: Supporting customer verification and account-opening processes.
Collections: Assisting with collection workflows and customer interactions.
Compliance: Supporting regulatory and internal compliance processes.
Fraud detection: Helping identify suspicious activity and potential risks.
Document processing: Applying AI to information extraction and banking documentation.
The companies also intend to explore applications through Pennant's existing banking agents, document-processing systems and collections intelligence solutions.
Pennant Founder and Managing Director Pradeep Varma explained the motivation behind the partnership:
"At Pennant, we're building agentic AI into our banking platform so banks and other regulated institutions can use AI across their operations while retaining control over their data and decisions."
This is an important point. Financial institutions do not simply need AI systems that generate impressive answers. They need systems that can operate within established controls and work alongside existing technology.
Why This Investment Matters for India's Sovereign AI Ambitions
For Bajaj Finserv, the investment is also connected to a wider view of how AI will influence critical sectors.
Chairman and Managing Director Sanjiv Bajaj highlighted the importance of building AI systems that are reliable, secure and auditable, particularly in financial services, cybersecurity and defence.
The partnership brings together three different capabilities: Soket's foundation-model research, Bajaj Finance's financial-services experience and Pennant's enterprise banking technology.
That combination could help move indigenous AI research closer to practical enterprise applications.
It also reflects a broader change in India's AI ecosystem. Financial institutions and technology companies are beginning to participate more directly in AI research rather than relying entirely on models and infrastructure developed elsewhere.
However, the investment does not mean that the planned banking agents are already deployed at scale. Their performance, security and ability to work reliably within regulated environments will need to be demonstrated.
The bottom line: The $4 million investment gives Soket AI strategic backing from two organisations with direct experience in financial services and enterprise technology. The next milestone will be turning its research into dependable AI applications for banking, where accuracy, security and control matter as much as intelligence.