I researched and explained how large language models (LLMs) work, describing training, attention, and fine-tuning.

I prepared and cleaned pharmaceutical documents using Python, applying data cleaning, JSON handling, text standardization, and image preprocessing across Colab notebooks.

I extracted text and bounding boxes from a 3-page SDF with PyMuPDF, reconstructed lines and tables, and produced structured field outputs for dates, lot numbers, and revision fields.

I compared Tesseract, PaddleOCR, and EasyOCR on a scanned pharmaceutical certificate and produced a documented engine-by-engine comparison with visualizations and recommendations.

I built and tuned a retrieval-augmented generation pipeline that extracted, chunked, embedded, and hybrid-retrieved document passages to produce grounded answers.

I compared Gemini, Mistral 7B, and Phi-2 in a consistent RAG workflow and reported quantitative and qualitative differences across accuracy, faithfulness, completeness, and speed.

I built a page-level metadata extractor and boundary detector that segmented a multi-document PDF and labeled each page by document type.

I built a Gradio PDF Q&A chatbot with multi‑file upload, TF‑IDF retrieval, and downloadable chat history.