What I'm doing right now

LEARNING

What I'm learning

• Continuing my undergrad in clinical medicine, keeping medicine as the foundation for understanding professional contexts and risk boundaries.
• Reading Designing Data-Intensive Applications to fill in the fundamentals of distributed data, reliability, and systems design.
• Keeping up with long-form AI interviews, focusing on how practitioners judge technical shifts, product opportunities, and engineering trade-offs — not just chasing release news.
• Catching up on Agent context, memory, tool calling, state management, and evaluation methods — beyond just the prompt level.
• Taking the judgments from books and interviews back into real projects, using code reading, testing, deployment, and incident post-mortems to check whether I really understand.

BUILDING

What I'm working on

• Continuing to maintain and run retrospectives on real business workflows: AI customer service, data sync, and business-data automation.
• Cataloging Coding Agent failure modes, turning hands-on impressions into checkable, reproducible tasks and verification points.
• Just wrapped my first round of upstream open-source contributions: three PRs adding medical modules to Academic Research Skills (40k+ stars) were all merged in early August. Still digesting what multiple rounds of adversarial review taught me.
• Maintaining this personal site, the open-source memory project, and this resume system with role-specific editing and export.

QUESTIONS

What I keep thinking about

• When has an Agent truly completed a business task, instead of just producing a plausible-sounding answer?
• How do we make failure alerts, result read-back, and human takeover default product capabilities?
• In high-risk settings like healthcare, how do we clearly separate knowledge assistance, product recommendations, and professional judgment?

INPUTS

How I process inputs

• DDIA gives me relatively stable systems knowledge; interviews give first-hand judgment from a fast-moving field. The two cross-check each other.
• I don't mistake "I've read it" for "I've learned it": I pull out the concepts, ask about the conditions where they apply, then try mapping them onto my own customer-service, data, and Agent projects.
• I plan to gradually turn the worthwhile reading notes and interview takeaways into blog posts, so the inputs leave checkable traces.

GAPS

Shortcomings I'm fixing

• I don't have a CS degree — complex architecture, security, and high-concurrency skills still need systematic study and professional collaboration.
• Running medical studies and AI projects in parallel means I need steadier long-term commitment, not just short-term sprints.
• Some product judgments come from personal use and real-project observation — they still need more rigorous sampling and evaluation design.

PRIVATE · ENCRYPTED

My goals

This part is more personal and keeps changing. It's encrypted — you'll need a password to read it.

Enter password to view →