Artificial intelligence is rapidly reshaping industries worldwide, and finance is no exception. Technologies such as machine learning (ML), large language models (LLMs), and natural language processing (NLP) are helping institutions process large volumes of structured and unstructured data, identify signals faster, and make more informed decisions. As financial markets become more information-driven and time-sensitive, AI is increasingly shifting from a research advantage to an operational necessity.
Jiazhao Shi, a software engineer at a leading global technology company and an independent applied AI researcher, has observed this shift through both engineering practice and academic research. Alongside his full-time role, he conducts independent research across transportation, fintech, and healthcare-related AI, focusing on systems that are both technically rigorous and practically useful. He has also served as a hackathon judge, mentoring early-career builders and supporting the next generation of engineers in AI and software development.
A Closer Look at FinSentLLM: Making Financial Sentiment AI More Practical
One of Jiazhao Shi’s recent fintech research contributions is the co-authored paper FinSentLLM: Multi-LLM and Structured Semantic Signals for Enhanced Financial Sentiment Forecasting. At a high level, the paper explores a practical question: can financial sentiment forecasting become more reliable by combining multiple AI models rather than relying on a single one?
The study introduces a lightweight framework that integrates multiple LLM “experts” with structured financial semantic signals, then uses a compact meta-classifier to generate the final sentiment prediction. It reports improved performance over strong baselines on the Financial PhraseBank benchmark and also examines whether the resulting sentiment signals are meaningfully related to real market behavior.
In simpler terms, the idea is similar to how financial institutions often rely on multiple analysts instead of a single opinion. Different LLMs may interpret tone, risk language, or market implications differently. FinSentLLM combines those perspectives and adds structured financial cues, making the final output more integrated and finance-aware rather than just the view of one model.
What makes this work especially relevant to fintech is that it goes beyond standard benchmark accuracy. Many sentiment studies stop at classification performance, but FinSentLLM also evaluates whether the generated signals have a meaningful long-run relationship with market dynamics through econometric analysis. In finance, that matters because institutions care not only about labeling text correctly, but also about whether model outputs are useful for market understanding, decision support, and downstream analytics.
More broadly, the work points to an important direction for fintech AI: progress may come less from a single “super model” and more from well-designed systems that combine language intelligence, domain-specific signals, and practical evaluation. In that sense, FinSentLLM reflects a broader shift toward AI pipelines that are more modular, interpretable, and aligned with real financial use cases.
Jiazhao Shi’s View on How AI Is Expanding Financial Intelligence
Jiazhao Shi sees today’s AI wave in finance as an expansion of financial intelligence, not a replacement for traditional expertise. In his view, AI creates the most value when it helps institutions combine signals from news, reports, market narratives, and structured financial indicators, so analysts can make faster, better-informed decisions.
He believes a key shift is the move from single-model pipelines to modular AI systems. Instead of relying on one model to do everything, financial teams are increasingly combining specialized components—LLMs for text understanding, structured signal engineering for domain context, and downstream models for integration and forecasting. He argues this approach is often more practical for enterprise adoption because it improves controllability, auditability, and tuning.
He also stresses that finance demands a higher evaluation standard than many consumer AI use cases. In finance, models must be robust, consistent, and aligned with domain needs, especially when outputs may affect risk assessment or strategic decisions. That is why he focuses on methods that not only improve accuracy, but also better capture financially meaningful signals and market behavior.
Looking Ahead: LLMs in Finance and the Next Phase of Adoption
Looking ahead, Jiazhao Shi believes LLMs will remain important in finance, but their value will come more from integration into broader financial intelligence systems than from standalone prompting. He expects the next phase of adoption to focus on hybrid architectures that combine LLM reasoning, structured financial data, domain-specific signals, and stronger evaluation frameworks.
He also sees growing demand for transparent, dependable AI systems. As institutions scale AI across research, risk, compliance, and investment workflows, success will depend not only on model capability, but also on governance, reproducibility, latency, cost efficiency, and explainability. In short, the future of AI in finance is not just stronger models, but more trustworthy systems.
For Jiazhao Shi, the biggest opportunities will go to researchers and engineers who can bridge academic innovation and real-world deployment, building AI tools that are both advanced and practical for high-stakes financial settings.
