AI/LLM Engineer Needed to Improve Accuracy of Existing AML Enforcement Chatbot

We have developed an AI-powered chatbot for a client in the regulatory compliance domain. The chatbot accepts user questions in natural language and returns curated responses based on data stored in a Convex database and extracted content from approximately 750 regulatory enforcement records and associated PDF documents. The system is already operational and produces accurate results for many queries. However, certain responses do not consistently meet the client's expectations, and we are looking for an experienced AI/LLM engineer to improve the accuracy and reliability of the solution. Scope of Work The selected freelancer will be responsible for analyzing the existing implementation and refining the query interpretation and retrieval logic to improve response quality. Key responsibilities include: * Reviewing and understanding the existing chatbot architecture and retrieval workflow. * Writing and modifying Convex functions used for data retrieval and aggregation. * Improving or redesigning portions of the LLM-based query interpretation layer that converts natural language questions into structured intents and actions. * Enhancing the logic that determines which Convex functions should be invoked for different user requests. * Implementing additional filtering and interpretation rules based on a detailed list of client requirements. * Identifying and resolving edge cases where the chatbot currently produces inaccurate or incomplete answers. * Testing and validating improvements against real-world user questions. Required Experience Applicants should demonstrate strong experience in: * Large Language Models (OpenAI, Anthropic Claude, Gemini, etc.) * Retrieval-Augmented Generation (RAG) systems * Prompt engineering and structured output generation * Intent classification and query planning * Convex database development * TypeScript / JavaScript * Designing AI systems that combine LLM reasoning with deterministic database queries Deliverables * Improved query interpretation and retrieval logic * Updated Convex functions (where required) * Enhanced filtering and aggregation logic * Documentation of key changes and recommendations * Demonstrated improvement in chatbot response accuracy Project Duration We estimate that the entire engagement should require no more than 30-40 hours, including analysis, implementation, testing, and documentation. Communication Requirements * Ability to work independently with minimal supervision. * Participation in two status calls per week, one of which will typically be scheduled on a weekend (preferably in the evening). * Clear communication and progress updates throughout the project. Application Requirements When applying, please provide: 1. A brief summary of your experience with LLM-based applications. 2. Details of at least one similar AI chatbot, RAG, or query-planning project completed within the last 12 months. 3. A short explanation of how you would approach improving the accuracy of an existing AI-powered retrieval system. Preference will be given to candidates who can demonstrate practical experience improving the accuracy and reliability of production AI systems rather than building simple prompt-based chatbots.