Fix ChatGPT Integration Accuracy

I have an AI-powered mobile app that connects to ChatGPT through the OpenAI API. Technically the calls succeed, yet the replies miss the mark: the model loses context and often returns information that isn’t relevant to the user’s prompt. This happens every time, so I know the problem sits in our integration or prompt strategy rather than in sporadic network glitches.

What I need from you is a surgical review and repair of the ChatGPT layer—everything from how we build and send the prompt, to how we handle conversation state, to any post-processing that might be muddying the output. If the solution calls for better prompt engineering, system-role framing, vector search, or even fine-tuning, I’m open to it as long as the end result is reliably accurate responses that understand context and surface the right information.

Please be ready to:

• walk through the current code (JavaScript/TypeScript with node-fetch and Express)
• diagnose the root cause of poor context retention and irrelevant answers
• implement and test the fix in a way that can be measured with repeatable prompts and sample conversations

I’ll spin up the dev environment for you, share the repo, and stay available for quick feedback cycles. Once the responses consistently reflect proper context and relevant facts across our test suite, we’re done.