DocMindAI

Project overview

DocMindAI turns PDF documents into a searchable knowledge base. It retrieves relevant passages through semantic search and supplies that context to a local language model so answers remain connected to the source document.

The problem

Long documents are difficult to search when readers need a precise answer, its supporting context, and the page where that information appeared.

Approach

1. 1

   Load uploaded PDF documents and split their text into searchable chunks.

2. 2

   Create sentence-transformer embeddings and store them in ChromaDB.

3. 3

   Retrieve semantically relevant passages for each natural-language question.

4. 4

   Generate a grounded response with a locally running Llama 3.1 model through Ollama.

5. 5

   Return source-document and page references alongside retrieved information.

Outcomes and insights

• Demonstrates a complete retrieval-augmented generation pipeline.

• Keeps language-model execution local rather than depending on a hosted model API.

• Provides a modular Python architecture for loading, chunking, indexing, retrieval, and question answering.