Lightning Talks + Closing Remarks DAY 1 | PyData Amsterdam 2024

Missing or biased data in healthcare and medical research can have fatal consequences, especially for women and underrepresented groups since most data is collected from male subjects

Time zone handling in data pipelines remains challenging - inconsistencies between UTC, local time zones, and daylight savings time can cause data duplication and analysis issues

Converting documents (like PDFs) to markdown using multimodal LLMs is becoming a viable alternative to traditional OCR approaches, with better accuracy and structure preservation

In NLP, the rapid evolution of models and techniques (BERT, GPT, LoRA, etc.) creates a steep learning curve for newcomers - understanding core concepts is more important than chasing latest trends

Clinical trial data management requires special consideration for security, standardization and proper handling of sensitive patient information

When working with data frames and time series, understanding the underlying data structure and intent is more important than simply choosing pandas by default

Data roles continue to evolve and specialize - from data analysts and scientists to ML engineers, research scientists, and data engineers, each with distinct skillsets

Test data generation and synthetic data creation remain challenging, especially for complex scenarios and edge cases

Code readability and maintainability should prioritize clear intent and documentation over minimal line count

Proper data visualization and communication are critical for conveying insights and findings effectively across teams