I explained how large language models process text, from tokenization through transformers to next-token prediction.

I cleaned a noisy scanned document using Python image preprocessing and produced a clearer image suitable for OCR comparison.

I extracted text and bounding boxes from multi-page mortgage PDFs, then applied heuristics to structure key fields like loan program, interest rate, loan term, and origination charges.

I compared Tesseract, PaddleOCR, and EasyOCR on mortgage PDFs and reported extraction issues and strengths for each engine.

I built and tested a complete RAG pipeline for document retrieval using LlamaIndex, producing AI answers extracted from mortgage-related PDFs and documenting the system design and results.

I tested and compared open embedding models to pick the best fit for my RAG pipeline.

I built a page-level segmentation and metadata extractor for multi-document PDF blobs.

I built a Gradio-based resume chatbot interface that parsed uploaded PDFs, surfaced source excerpts, and answered recruitment questions via preconfigured prompts.

I reviewed a 39-page pharmaceutical document packet and cataloged its regulatory file types, date formats, and parsing challenges.

I prepared and cleaned pharmaceutical-style documents using Python in Google Colab, applying data cleaning, JSON handling, text standardization, and image preprocessing to improve OCR readiness.

I extracted and structured text from sample SDF PDFs using PyMuPDF and reported findings for field extraction.

I compared Tesseract, PaddleOCR, and EasyOCR on scanned pharmaceutical supplier PDFs and documented the extraction results.

I built a complete RAG pipeline using LlamaIndex in Colab, documented architecture choices, and produced a demo and write-up.

I ran a side-by-side evaluation of Gemini, Mistral-7B, and Phi-2 within RAG pipelines and documented differences in reliability, latency, and response quality.