lfoppiano / document-qa

Scientific Document Insight Q/A

Hi there 👋

Artificial intelligence specialist with 10+ years of experience in software engineering. I have expertise in Text and Data Mining, Natural Language Processing, and Data Science. I'm interested in the development of specialized processes for scientific text, in particular document parsing and structuring.

I like mostly anything related to the outdoor and travel.

document-qa's Issues

improve interface, makes the keys not mandatory whey they are provided as env var

Remove documentation from side bar and move there the settings

Add documentation on the document-qa-engine library

Add documentation on how to use the document-qa-engine library

add orca

Replace mistral with Orca, which need a paid API handler...

https://huggingface.co/microsoft/Orca-2-13b

Display an error message when grobid is not responding

From #8 it seems that if the grobid service does not reply on time, we get some unclear behaviour.

Need to verify the error code and show a message saying "please try later"

Set custom context

Idea: use context from highlighted text from the PDF document

Streamlit-pdf-viewer Component Interfering with Chat History Auto-Scroll Functionality

Description

I've encountered an issue when implementing the streamlit-pdf-viewer component in the document-qa application. It appears that the auto-scroll functionality of the chat history, which normally brings the view to the bottom of the chat when a user gets an answer from generative AI, stops working.

Potential Cause and Solution

I suspect this might be related to the current limitation of the st.chat_input widget, which cannot be used inside st.columns. If st.chat_input were adaptable for use within st.columns, this might resolve the issue.

Relevant Information

According to a recent comment on issue #7296 in the Streamlit repository, there is a planned implementation to solve this problem within the next three months.

Current Approach

For now, the strategy is to wait for updates or fixes to the Streamlit library.

Load the PDF file in the interface

We can load the PDF in the interface, on the left side, leaving the chat on the right side.

add chat memory

• add memory between each chat messages and responses
• at every document uploaded the memoery should be reset

Example:

interact with the pdf document loaded

We could link the paragraphs extracted from the RAG with the PDF coordinates and show some visual aids of the part of the document used for answering...

Add summarization

To properly summarise papers, we need to iteratively compress information in a slightly different way than the Q/A.

The summarisation could be implemented as separate function than the Q/A, in the same interface.

Could not browse (read) the uploaded pdf file

Environment: Safari Version 17.0 (19616.1.27.211.1)
Frequency: every time
Steps to reproduce error:

1. Input the Chat GPT API-key
2. Browse file from local source (pdf)
3. Error happened when executed

TypeError: This app has encountered an error. The original error message is redacted to prevent data leaks. Full error details have been recorded in the logs (if you're on Streamlit Cloud, click on 'Manage app' in the lower right of your app).
Traceback:
File "/home/appuser/venv/lib/python3.10/site-packages/streamlit/runtime/scriptrunner/script_runner.py", line 541, in _run_script
    exec(code, module.__dict__)
File "/app/document-qa/streamlit_app.py", line 262, in <module>
    st.session_state['doc_id'] = hash = st.session_state['rqa'][model].create_memory_embeddings(tmp_file.name,
File "/app/document-qa/document_qa/document_qa_engine.py", line 204, in create_memory_embeddings
    texts, metadata, ids = self.get_text_from_document(pdf_path, chunk_size=chunk_size, perc_overlap=perc_overlap)
File "/app/document-qa/document_qa/document_qa_engine.py", line 169, in get_text_from_document

Add NER highlight on responses from the system

• grobid-quantities
• grobid-superconductors

Load different structures

We can extract three type of structures, which are then loaded in separate storages:

• fulltext
• figures / tables
• bibliographic data
• references / text callout