RAG / LLM Prompt Engineer to Audit & Improve Document-Based AI Assistant (OpenAI + Qdrant)

I’m building an AI assistant that allows users to chat with uploaded documents.

Current setup

• Document upload (PDF)
• Optional structured JSON extraction
• Chunking + embeddings
• Stored in Qdrant
• Retrieval per query
• OpenAI (ChatGPT) generates responses

The system works, but responses are sometimes shallow, inconsistent, or not strongly grounded in the document.

I’m looking for an experienced RAG / LLM engineer to:

• Review the architecture
• Evaluate chunking and retrieval strategy
• Review prompt structure (system + context + conversation)
• Assess whether structured extraction is necessary
• Improve grounding and response quality
• Clearly explain what should change and why

This is an architectural review and optimization project, not a build-from-scratch task.

Tech stack

• Node.js
• OpenAI API
• Qdrant vector database

Please include

• Your experience with RAG systems
• Similar systems you’ve improved
• How would you approach auditing this

Strong English communication required.