𝐒𝐨𝐥𝐮𝐭𝐢𝐨𝐧丨Improving LLM Math Reasoning with Prompting and RAG

AI Prompt Engineer & Algorithm Developer

𝐒𝐜𝐨𝐩𝐞 & 𝐂𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞: Tasked with improving mathematical reasoning in an instructed LLM. The core challenge was overcoming a low baseline accuracy and frequent reasoning errors caused by missing contextual knowledge.

𝐃𝐞𝐥𝐢𝐯𝐞𝐫𝐚𝐛𝐥𝐞𝐬 & 𝐅𝐞𝐞𝐝𝐛𝐚𝐜𝐤: Engineered a two-stage pipeline using structured Chain-of-Thought (CoT) prompting and RAG to dynamically inject theorems. This reduced errors and boosted accuracy from 60% to 85%. The workflow now powers the client's B2B projects, securing an ongoing long-term technical partnership.