Query Engine Integration : Integrating multiple index types (summary, vector, and knowledge graph) into a unified query engine was challenging. The solution involved developing a custom QueryEngineBuilder class that combines these indexes using a RouterQueryEngine, allowing for more comprehensive and accurate responses.

Memory Management: As the application processed multiple requests, memory usage issues were encountered. This was addressed by implementing a document store that saves processed information to disk and loads it on-demand, reducing memory overhead.

Asynchronous Processing: Handling document uploads and processing asynchronously while providing immediate feedback to users was challenging. This was addressed by using FastAPI's background tasks for document processing.

LLM Halucination: The LLM starts hallucinating after certain conversations or couldnt extract keyfindings with large commits , this was overcome by using proper and structured prompts with example and negatives mentioned

Responsive Design: Ensuring that complex UI elements, such as the chat interface and document analysis results, were responsive across different device sizes presented layout and usability challenges.

State Management: Managing the application state, particularly for the chat history and document analysis results, across different components and pages in Next.js was complex. This has been addressed using state management libraries or Next.js's built-in solutions.

Error Handling and User Feedback: Implementing comprehensive error handling on both the frontend and backend, and providing meaningful feedback to users, especially during long-running operations like document analysis, was challenging.

Deployment Problem: Implementing an embedding model locally caused the Cloud instance to overflow memory. This was overcome by using GoogleGenerativeAIEmbeddings