The LangChain library provides tools to interact with SQL databases and build queries based on natural language inputs. However, building a custom solution that leverages domain-specific knowledge can improve the agent's performance and efficiency. The standard SQL Toolkit has limitations when it comes to handling complex databases, leading to problems such as incorrect query generation, excessive tool usage, and irrelevant prompts. Extending the toolkit with innovative RAG techniques can overcome these issues by incorporating domain-specific knowledge into the prompt template. This approach involves retrieving relevant data from a vector database using Retrieval Augmented Generation (RAG) and dynamically including it in the prompt to improve query generation accuracy. Additionally, applying few-shot examples and making the system robust to misspellings can further enhance the agent's performance. By leveraging developer field-specific knowledge and exploring new techniques such as similarity thresholds and prioritizing diversity of few-shot examples, the standard SQL Toolkit can be improved to provide a more efficient and accurate solution for building custom LLM-SQL solutions.