The AI landscape is currently in a rapid evolution vortex. There would be a lot more AI based use-cases that will solve real world problems with the help of various LLMs. Ravindra Nagpurkar, Founding Partner, TheSybil.ai explains to Bhavya Bagga, Business Reporter (Corporate & Leadership), CXO Media & APAC Media exclusively that the AI surface has barely been scratched though agentic AI can now make sentimental and behavioral analysis a possibility.
What inspired the founding of TheSybil.ai, and what core technological challenges is the platform aiming to solve across industries?
Sybil.AI is the first product to come out of our AI labs, which has been established to build AI first solutions for enterprise solutions.
The motivation to build an AI based and ML backed product for the capital markets was multifold. This industry’s inherent DNA is based on vastness and complexity of data. Add to that the need for speed and efficiency. Traders demand real time insights and equally rapid decision making.
All of this is a function of data and with the advent of multiple ML models and AI’s ability to draw insights makes decision making almost automatic and free of human emotions and biases. This is a huge factor in determining outcome.
Agentic AI makes sentimental and behavioral analysis now a possibility and one can track it real time on our platform. This is the barometer of market participants and shows a distinct forward-looking trend for our clients to follow.
In addition, the advent of automation tools coupled with the data fabric makes automated risk management and trade execution very much a part of the trading ethos. We use this for managing our AUM and also license this technology to HNI clients.
As CTO, how are you leveraging AI and machine learning to build scalable, secure, and context-aware enterprise solutions at TheSybil.ai?
To leverage AI and machine learning involves several key strategies:
- Scalability
Distributed Computing: We utilize cloud solutions provided by AWS to deploy and scale ML models in real time to growing data volumes and data. A microservices driven design and architecture breaks down solutions into modular, independent components that can be scaled horizontally.
Model Optimization: Model pruning, quantization and efficient architectures (like transformer variants) help to enable faster inference while keeping resource consumption under check and within defined bounds.
- Security
Model Security: We have deployed a mix of model security techniques like adversarial testing, monitoring for model drift and secure model deployment practices to prevent unauthorized access or data tampering
- Context-Aware Capabilities
Real-Time Data Integration: Live data streams from multiple sources are used to enrich the models to provide latest and real time context
Personalization: Market participants profiles and their behavioral data is used to tailor insights and recommendations for a cross section of users. Each user can potentially have his/her own context based on the trading strategies and products they use.
By integrating these strategies, we create AI-driven solutions that are not only powerful and scalable but also secure and personalized to the specific, nuanced context of various types of users. This ensures reliability and relevance in fast-paced and sensitive domains like finance and capital markets.
Data privacy and security remain paramount in AI-led platforms. How is TheSybil.ai addressing regulatory compliance, ethical AI usage, and data protection in its architecture?
Data privacy, security, and ethical AI usage are the cornerstones of any AI platform and addressing them effectively requires both strategic governance and robust architectural measures.
We make use of some of the privacy and security architecture principles such as
Privacy by Design & Secure by Design
The foundational principles of privacy by design such as privacy as default and user centricity combined with secure by design principles such as embedding security features like encryption at every stage makes the platform secure and comply with various technical regulations.
Zero Trust Architecture & Role-Based Access
We have adapted a Zero Trust model—”never trust, always verify,” verifying every user and device before granting access.
Confidential Computing / Secure Enclaves
By leveraging hardware-based trusted execution technologies such as SGX and SEV to process sensitive data in protected environments, even cloud providers can’t access this data. This is paramount for secure model training, inference, and personalization.
By combining this ideology with privacy-first, secure architecture and an ethical culture, Sibil.AI ensures adhering to regulatory compliance framework and safeguards user data by upholding ethical principles
How do you approach cross-functional collaboration between engineering, product, and business teams when developing AI-driven solutions that must meet both technical and real-world use case demands?
Cross-functional collaboration is essential for successful development of an AI-driven solution that is technically robust and aligned with real-world business objectives. Here’s a proven approach we implement to bridge engineering, product, and business functions effectively:
Align Early on the “Why”
Before building anything, get all teams to agree on:
- The business problem being solved
- The expected value of the AI solution
- The success criteria (OKRs)
Build Cross-Functional Squads or Pods
Build Integrated Teams: Form small teams that include representatives from:
- Engineering (data + platform + ML)
- Product management
- Business (e.g., marketing, operations, or customer success)
And we champion the Agile development framework upon which all our product development roadmap is based and executed.
What does your current technology stack look like, and what factors do you prioritize when evaluating emerging tools, platforms, or infrastructure providers?
A typical AI/ML tech stack spans several layers from data ingestion to model deployment, and varies by scale, maturity, and use case. Some of the technologies we use include Apachi Kafka messages for data ingestion, dbt and PySpark for transformation and Vertex AI for business features.
For model development and deployment, we use PyTorch and MLFlowwhile for computation we rely on Vertix AI and sometimes AWS SageMaker.
Factors for evaluation of these technologies are broadly dependent on scalability, Security, Hardware Cost efficiency (Cloud) and Observability.
Looking ahead, what are your strategic priorities for TheSybil.ai over the next 12–18 months, and how do you see the broader AI landscape evolving in India and globally?
This AI product is one of many that we will build as part of our AI Labs mission. In the coming 2 years you will hear about multiple solutions for various enterprise verticals in the financial and non-financial domain.
It is too early to speak about that product roadmap as we are constantly experimenting with multiple AI use-cases and adapting to the new LLMs that are now available for all to build on top of.
The AI landscape is currently in a rapid evolution vortex and what is clear is there would be a lot more AI based use-cases that will solve real world problems with the help of various LLMs (General and Niche trained). We have barely scratched the surface.
What is clear is lean teams will finally be possible and efficiencies at scale will make an impact at a much larger scale than automation did in the past decade.
