Hi everyone. I'm honored to be here. My name is Kiran Otti, and today I'll be talking about swarming LLM agents for real time market insight in cloud native trading systems. This talk brings together three key areas, financial markets, large language models, and Kubernetes native architectures. My goal is to show you how cloud native swarming agents can transform the way we process and act on financial data in real time. Okay, let's start with the challenge. Financial markets generate more than 2.5 quintillion bytes of data every single day. What's more striking is that nearly 80% of this data is unstructured. Think about news articles, social media chatter, analyst reports, or SCC filings. Traditional models struggle here. They can't keep up with the scale. They are, don't adapt well to complexity, and the latency of analysis often means. Opportunities are lost in trading. Seconds matter and missing signals buried in noise translates directly into missed opportunities. To address these challenges we developed self adapting financial sentiment, Oracles. It's built as a cloud native framework designed to ingest massive streams of unstructured financial data in real time. It uses warm intelligence, meaning multi LLM agents. Each with its own specialization working together to generate insights. Think of it as a team of experts, collaborating instead of single generalist agent, and importantly, the self adapting financial sentiment. Oracle provides privacy, preserving, federated learning, powered by homomophic encryption. This allows institutions to collaborate without ever sharing sensitive raw data a must in today's regulatory environment. Okay. Now looking at the technical architecture, at the heart of the self adapting financial sentiment, Oracles is the dual model architecture. Finberg is used to extract precise domain specific sentiment from financial texts, such as earnings calls and reports. Robot large complement it by providing broad contextual understanding from market news and social media. The system runs this containerized microservices orchestrated by Kubernetes. HPA handles dynamic scaling while Kubernetes. A self-healing ensures reliability. As a result, we continue, we consistently achieve sub fortify millisecond latency even during extreme market volatility. The agent swarm itself is structured like this. We have the news agents which analyzes breaking news and press releases. We have the social media sentiment analysis agents. Tracking investor conversations across social media and different other forums, regulatory filing agents passes, SEC, filings and disclosures for critical updates. Market data agents link the sentiment to actual price, action and order flow. Each agent is domain specific, like I mentioned, but together they provide a rich multidimensional view of the market. Because of this distributed architecture, the self adapting financial sentiment, Oracles processes around 1200 documents every second. Equally important, the end-to-end latency from ingestion to insight is still under fortify milliseconds in the world of high frequency trading where every millisecond counts. This combination of throughput and speed is really transformative. But how do we bring all of these agents together? That's the role of the consensus. Oracle, it uses a 16 head attention mechanism to weigh predictions from different agents. It applies dynamic confidence scoring to quantify uncertainty, and it continuously self calibrates as the market conditions shift. This means our predictions are not only fasted, but also robust and adaptive in nature. We've also implemented these self adapting financial sentiment, oracles across 47 financial institutions. Each institution's train models locally on their private data. Instead of sharing raw data, only encrypted model weights as sent back to the central orchestrator. This federated setup improved accuracy by almost 15% compared to isolated models, all while meeting the strictest of privacy and compliance requirements. In terms of the performance metrics, here's what the system is able to achieve. We are able to achieve around 95% directional accuracy, predicting whether the market goes up or down. We all also achieved 87% magnitude accuracy, estimating how the markets move. 21% analyzed returns over five years of back testing a sharper ratio of 1.67. A strong measure of risk adjusted performance. In short, the self adapting. Financial sentiment. Oracles is in just theoretical, it consistently outperforms traditional sentiment systems. All of this is, of course, made possible by cloud native design. Each agent runs in a lightweight container, auto-scaling, adjust to incoming data surges and STO service mesh provides a secure, resilient interservice communication. And Prometheus monitoring is used that delivers real time health and performance metrics. It's a Kubernetes first design ensuring both scalability and resiliency beyond just prediction, self adapting financial sentiment. Oracles also act as an early warring system. It can identify market stress events up to 36, 36 hours in advance with over 96, 90 2% precision. It also detects coordinated manipulation patterns. For example, when the social media campaign attempts to influence the market sentiment, this makes it powerful tool not just for profit, but also for risk management purposes. Of course, it wasn't easy. We faced four key challenges, latency management, balancing model complexity with speed, data privacy and compliance. Meeting the G-D-P-R-C-C-P-A and other regulations. The model drift, keeping models accurate as the data was. And resource optimization when it comes to tuning Kubernetes clusters to balance out cost with performance, these challenges pushed us to innovate not just in models, but also in the infrastructure space. In terms of looking ahead for the future development roadmap, we are expanding the self adapting financial sentiment. Oracles with multi-language support, cross asset correlation, audio sentiment analysis from earnings calls. ESG factor integration and few other innovative features. The goal is to make the sentiment the self adapting financial sentiment. Oracles not only smarter, but also future proof. To summarize swarming LLM agents outperform single monolithic models. The cloud native architecture is essential for the scale and speed of modern trading. Federated learning delivers accuracy while protecting privacy. And the consensus mechanisms ensure more reliable predictive insights. Together these elements provide the speed, intelligence, and trustworthiness needed for the next generation trading system. Thank you all. Please let me know if there are any questions.