2. Related Work

Recent research has demonstrated the growing capability of AI systems in financial prediction:

• Specialized financial language models such as FinGPT (Yang et al., 2023) have shown that domain-specific training and data-centric approaches significantly improve performance on financial tasks compared to general-purpose LLMs.
• Comprehensive surveys, including “The New Quant: A Survey of Large Language Models in Financial Prediction and Trading” (Fu, 2025), highlight the shift toward hybrid systems that combine numerical time-series modeling, retrieval-augmented generation, and agentic reasoning.
• Studies on LLM-based return prediction (e.g., works examining GPT-4’s ability to extract directional signals from news and market data) confirm that language models can capture economically meaningful information, particularly when grounded in structured market inputs.
• Multi-agent and hybrid architectures have shown improved robustness in volatile environments such as cryptocurrency markets.

ZKSeer builds on these findings by moving beyond prompt-based or lightly fine-tuned general models. Instead, it deploys purpose-built forecasting models trained specifically on market microstructure and short-horizon price behavior.