How to bypass aggressive forced CAPTCHA on a dynamic academic repository (CUHK)?

Hi everyone,

I am building an automated scraper for an academic repository (https://repository.lib.cuhk.edu.hk/en/collection/etd/year), which structures resources by year. However, we've hit a hard block.

Here is the situation:

1. Aggressive CAPTCHA: The site uses a mandatory CAPTCHA (image verification) system that triggers for almost every visit, regardless of whether it's a real human browser or a bot.
2. AI/Agent Failure: Our automated agents attempted to solve the image verification multiple times but were consistently blocked.
3. JS-Heavy: The site is heavily reliant on JavaScript to load the actual item lists, making simple HTTP requests (requests/curl) useless.

My Questions:

1. What is the state-of-the-art stack for bypassing such aggressive image CAPTCHAs in a fully automated Python pipeline today? (e.g., combining undetected-chromedriver with a specific solver service?)
2. Since this is an academic repository, it likely supports OAI-PMH. Does anyone have experience bypassing the frontend completely by finding standard API endpoints (like /oai/request) on similar library systems?
3. Any specific configuration recommendations for browser automation to reduce the CAPTCHA difficulty/frequency on this specific type of site?

Thanks in advance for any insights!