/llm-agent-gmail_parser-better-rag

Prompt techiniques to parse emails to RAG applications

Primary LanguageJupyter Notebook

LLM Agent Gmail Parser Better RAG 📧🤖

Problem Addressed

Indexing emails to use with RAG applications or to extract relevant knowledge has its challenges, such as:

  • A lot of noise and garbage in the text
  • Dealing with email threads

Solution

In this project, I experimented with various prompt techniques to extract the most important information from emails while addressing the problems stated above. I found that a report-style summary works better than just asking for a summary, as the latter tends to lose a lot of important information.

Key Points 🌟

  • Noise Reduction: Implemented techniques to filter out irrelevant information and focus on key content.
  • Thread Handling: Developed methods to accurately parse and summarize email threads.
  • Report-Style Summaries: Discovered that report-style summaries retain more essential information compared to generic summaries.
  • Prompt Engineering: Experimented with different prompt structures to enhance the extraction of valuable insights.