Financial Prompt Engineering Specialist

I'm seeking an experienced prompt engineer with expertise in equity research, broker reports, and financial terminology. Your task is to enhance a GPT-based prompt system to improve the classification of financial data.
We have a prompt that:
Ingests broker reports (PDF or text)
Applies Chain-of-Thought (CoT) logic
Classifies a particular stock as Positive / Neutral / Negative based on:
Target Price
CMP
Analyst tone
Valuation metrics
Forward estimates, etc.

Problem Statement:
When running the prompt on more than 2-3 reports in the same thread, GPT starts:
Mixing up content from multiple reports
Confusing company names, target prices, and CMPs
Generating incorrect or random outputs due to memory bleed across turns

We want to fix this so that:
The prompt reliably processes one report at a time
Each report’s context is isolated
Classification outputs are accurate and explainable

Your Role:
Review the current prompt and workflow
Refactor it using robust prompt engineering principles
Ensure report-level isolation, context control, and reliable parsing
Add error checks or fallback logic where needed

Requirements:
Proven experience with ChatGPT, GPT-4/4o, or similar LLMs
Solid understanding of equity research reports and stock classification logic
Strong prompt engineering portfolio (CoT, RAG, context windows, etc.)
Ability to work with multi-document input handling

Deliverables
Refined prompt that avoids context mixing
Test cases showing reliable classification across multiple reports
Documentation of changes and logic applied