I explored Machine Learning, Deep Learning, LLMs, and NLP to understand how AI reads and processes pharmaceutical documents. Analyzing a real pharmaceutical documentation file, I gained insights into how AI models handle both structured and unstructured data, enhancing my foundation in AI's role in

I utilized Python for data processing and cleaning, preparing pharmaceutical documents for AI automation. By working with Google Colab, I learned the basics of Python and applied various data preparation techniques to ensure accurate and efficient document intelligence.

I transformed digital PDFs into machine-readable formats using Python, employing tools like PyMuPDF and pdfplumber to extract text from multi-page pharmaceutical SDFs. I implemented field heuristics to identify key fields, establishing a compliance workflow for flagging outdated documents.

I engaged with three OCR engines—Tesseract, PaddleOCR, and EasyOCR—to evaluate their effectiveness on scanned pharmaceutical supplier documentation. By comparing their performance, I recommended the most suitable tool for real-world automation, focusing on both technical and business needs.

I built a Retrieval-Augmented Generation (RAG) pipeline using LlamaIndex to enhance AI’s capability to retrieve relevant information from extensive document collections, addressing the challenge of quickly finding necessary data.

I improved the RAG pipeline by implementing advanced chunking techniques and metadata-based filtering for better document segmentation and retrieval accuracy. This optimization allowed for a more contextually relevant response from open-source models.

I designed a modular system to process large, unstructured pharmaceutical regulatory files, classifying documents and routing them through appropriate extraction pipelines. This project reflected real-world pharmaceutical document workflows.

I developed a Gradio chatbot that integrated with my local RAG pipeline, allowing users to interact with the system through a user-friendly web interface. This project enhanced the accessibility of my document intelligence solution.