Fix Hallucination Issues in Mira AI Recruitment Agent (LangChain + RAG)

Hey, I have a Python AI recruitment agent called Mira that’s been running well overall. She’s built with LangChain and uses a RAG pipeline to pull answers from a knowledge base, but there’s a reliability issue I need sorted: she occasionally hallucinates. Not constantly, but enough to matter. Sometimes Mira gives an answer that sounds reasonable but isn’t supported by anything in the retrieved documents. Other times she blends real retrieved content with things she’s generating from her own training data. In a recruitment setting that creates real problems, so I need her responses properly grounded in the knowledge base. When something isn’t there, she should say so rather than fill in the blank. The core app is functional. The pipeline runs, retrieval is working, the agent responds. This isn’t a rebuild, it’s a focused debugging task for someone who knows RAG systems well. What the work involves: You’d review Mira’s retrieval logic, how context gets passed into the prompt, and how the model is instructed to use it. Pinpoint the root cause, apply the fix, and verify it holds across a range of test inputs including edge cases where retrieval comes up empty, since that’s currently where most of the hallucination happens. Deliverables: Updated working code with the fix applied, plus a short write-up covering what caused the issue, what was changed, and any suggestions for keeping Mira reliable as the project scales. A few paragraphs is fine, nothing lengthy. Good fit if you have: Hands-on experience debugging LangChain and RAG pipelines, solid understanding of prompt engineering around context grounding, and familiarity with vector search and embedding strategies. Experience with OpenAI, Anthropic, or similar APIs is a plus. To apply: Share a brief note on relevant experience and a couple of examples of similar work. That’s all I need to move forward. I’m ready to start quickly and if this goes well there’s more work on Mira to follow. Alex