Mengsay Loem

Mengsay Loem · Follow 5 min read · -- Document digitization is crucial for streamlining workflows and reducing manual labor, particularly with many organizations holding vast archives of paper files. While classical OCR-based pipelines have historically been the backbone of automated document analysis, recent advancements in Vision Language Models (VLMs) present an intriguing alternative — especially when framed in a Visual Question Answering (VQA) context.

⚠️ WARNING The effectiveness of these tips may vary across different models and contexts. It's essential to understand the LLM's limitations and know that successful prompts may not work universally.

Transcripts. The main data needed are the commencement speech transcripts. After some googling, I decided on two sources. The first one is from a dataset collected and organized by Charlie Harrington on FloyHub, which can be downloaded directly. The second is from the website Graduation Wisdom, which can be scraped with BeautifulSoup. Upon merging data and dropping the duplicates, I ended up with 400+ transcripts. Speaker and school info.