I explained how large language models process text, from tokenization through pre-training and fine-tuning, and described the Transformer self-attention core that enables contextual predictions.

I prepared and processed pharmaceutical documents using Python in Google Colab, producing cleaned data and image-preprocessed files ready for OCR and downstream AI tasks.

I extracted and structured text and bounding boxes from multi-page SDF PDFs using PyMuPDF, then analyzed extraction gaps and proposed anchor-based field matching.

I compared Tesseract, PaddleOCR, and EasyOCR on a scanned pharmaceutical supplier PDF and reported which engine preserved layout and key fields best.

I built and tested a RAG pipeline for document retrieval using LlamaIndex, embeddings, chunking, hybrid retrieval, and reranking.

I ran and compared three RAG configurations on a sample SDF document, logging retrieved chunks, answers, and confidence to pick the best retriever settings.

I built a page-level metadata extractor that identified document boundaries and labeled each page by document type from bundled PDFs.

I built a Gradio-based chatbot interface and connected it to a local RAG pipeline, handling modes, multi-file uploads, and chat history features.