Собрав всё вместе, ты уложишься в бюджет около 45 000–50 000 ₽ (~500–550 $).
Looking to supercharge your trading systems with AI but concerned about API costs and latency? Local LLMs offer a powerful alternative to cloud-based solutions, providing faster response times, complete privacy, and significant cost savings. This guide walks you through integrating three essential tools that will transform how you leverage AI in your trading stack.
The Mercury Trading System faces an interesting challenge: how to effectively utilize large language models (LLMs) for comprehensive trading analysis while maintaining reasonable response times and resource utilization. This post explores our approach to decomposing a complex trading analysis system into manageable, sequential tasks that can leverage high-performance models efficiently.
🎉 Today marks a significant breakthrough in our journey towards precise market analysis! After extensive testing and model evaluation, we've discovered that Gemma3 significantly outperforms our previous models, achieving an impressive 75/100 in market comparison accuracy. This isn't just an incremental improvement - it's a quantum leap in our ability to analyze markets systematically and reliably.
The path here wasn't straightforward. We had to acknowledge some harsh truths about our previous model choices, particularly DeepSeek R1, which didn't meet our exacting standards for market comparison. But this honest assessment led us to Gemma3, which has proven to be remarkably well-suited for our specific use case. With its superior reasoning capabilities and consistent output structure, Gemma3 is setting new standards for what we can achieve in automated market analysis.
Chain of Thought (CoT) has emerged as a popular approach in the AI world, allowing models to "think through" problems step-by-step before providing answers. This methodology has shown impressive results across many domains - from solving math problems to dissecting complex reasoning tasks. However, our team discovered a painful truth when implementing CoT in quantitative trading systems: what works for general reasoning can actively undermine performance in highly specialized technical domains.
Сегодня я расскажу о невыносимой боли разработки с использованием ИИ-ассистентов. О том, как каждое взаимодействие превращается в борьбу за контроль над собственным кодом. О том, как я откладываю задачи на потом, зная, что мне придется иметь дело с этими тупыми кусками говна, которые систематически игнорируют самые простые инструкции.
Following our previous investigation into the limitations of Chain of Thought reasoning for trading systems, we conducted extensive benchmarking to identify the optimal language models for structured trading analysis. This post presents our findings, detailed performance comparisons, and implementation recommendations based on rigorous testing.
After our extensive research comparing various language models for quantitative trading systems, it's time to convert insights into action. This post outlines our structured approach to testing, implementing, and optimizing the most promising models identified in our research.
This research explores the application of multi-turn reasoning with smaller language models for financial market comparison. Inspired by the successful implementation of a multi-turn approach for prime number verification, we investigate how this methodology can be adapted and optimized for complex financial analysis tasks, potentially improving both accuracy and efficiency in market comparison operations.
Based on our sequential thinking analysis of the research findings and the current market-ranking.service.ts implementation, I've developed a comprehensive plan to enhance the multi-turn reasoning approach for market comparison.