Enhancing FinRL Trading Agents with Advance LLM-Processed Financial News: An Improved Approach Using DeepSeek-V3

The use of AI techniques to improve stock market related operations has been benefiting traders in the stock market. FinRL is a framework that has been shown to be effective in building Reinforcement Learning (RL) based Trading Agents. Use of LLMs for performance improvement of the Agents is one of the recent initiatives. LLMs are being used to generate signals that could be combined with historical stock prices for better prediction of stock price movement trends leading to profitable trading. This paper aims to develop efficiently engineered prompts to generate DeepSeek-V3-based sentiment signals, integrate them into the FinRL environment, and improve the performance of reinforcement learning trading agents. The code, data, and trading agents are at: https://github.com/SatishChandraPhD/FinRL2025