Fix Custom GPT for Data Analysis
Budget: $30 – $250 USD
My custom GPT should run data analysis on user behavior, but right now it keeps generating incorrect responses and often fails to pick up the surrounding context of a prompt. I need a concise diagnostic and quick repair that gets it back on track.
Scope of work
• Review the current prompt chain, system messages, and any fine-tuning or embedding steps I’m using.
• Identify why it loses context and produces wrong answers.
• Apply lightweight fixes—prompt engineering tweaks, parameter adjustments, or minor code corrections—to improve accuracy.
• Verify the model can parse user-behavior datasets and produce coherent, context-aware insights.
• Deliver a short summary (bullet-point or markdown) of the changes you made and why they work so I can replicate them later.
Tech notes
– The GPT runs on the OpenAI API inside a small Python script.
– Data arrives as cleaned JSON lines; Pandas and NumPy are available if you need them.
– No major redesign—just a focused patch and guidance.
If you’ve debugged prompt chains or tuned GPT-3.5/4 models for analytical tasks before, this should be a fast, straightforward engagement.
Scope of work
• Review the current prompt chain, system messages, and any fine-tuning or embedding steps I’m using.
• Identify why it loses context and produces wrong answers.
• Apply lightweight fixes—prompt engineering tweaks, parameter adjustments, or minor code corrections—to improve accuracy.
• Verify the model can parse user-behavior datasets and produce coherent, context-aware insights.
• Deliver a short summary (bullet-point or markdown) of the changes you made and why they work so I can replicate them later.
Tech notes
– The GPT runs on the OpenAI API inside a small Python script.
– Data arrives as cleaned JSON lines; Pandas and NumPy are available if you need them.
– No major redesign—just a focused patch and guidance.
If you’ve debugged prompt chains or tuned GPT-3.5/4 models for analytical tasks before, this should be a fast, straightforward engagement.