1 About Your Instructor
1.1 How I Got Here
Carlos Oya reached out after seeing a recent ODI Global paper I co-authored with Yunnan Chen called Greener on the other side? Mapping China’s overseas co-financing and financial innovation. We used a novel LLM-based approach to classify “green” Chinese lending projects—something that would have taken a large research team months to do manually.
When we did this work, I looked for “best practices” for validating LLM findings. There weren’t many. So we developed our own validation method and published both a methodological appendix and our GitHub repository. It’s not perfect, but it’s something for others to build upon.
- 18,000 projects classified in 15 hours using Deepseek v3
- $1.58 total cost vs. estimated $22,500 for manual classification
- 91.8% agreement with human raters on validation sample
- First comprehensive analysis of China’s green overseas lending portfolio
This experience showed me how LLMs can enhance what’s possible for policy-relevant research. Two policy researchers on a tight budget accomplished what traditionally required large, grant-funded research teams. There’s a long way to go to establish best practices for the use of LLMs in policy research, so I’m trying to do my best to move the conversation forward.
1.2 What I Do Now
Day job: Running Teal Insights, where we help Global South finance ministries navigate complex debt sustainability and climate investment challenges. We’re philanthropically funded with a mandate to build open-source tools—including LLM tools—so countries don’t have to pay exorbitant fees to financial advisors.
Our approach: Small team (US, Nigeria, Kenya) using AI tools heavily to amplify our impact in research and code development.
1.3 Other Relevant Experience
- EM sovereign debt resarch analyst, Morgan Stanley Investment Management
- Adjunct Lecturer, Johns Hopkins SAIS (teaching students to do real-world data analysis on financial and sustainability data)
- Thought leadership on sovereign debt + sustainability, World Bank
- Chinese debt restructuring & flows research, AidData
- Big nerd 🤓
1.4 Why This Workshop?
This technology is very new. Nobody is really an “expert” yet. But since we’re using these tools extensively, we’ve learned hard lessons about how to use them well—and badly.
When Carlos asked me to teach this, I figured it was a great excuse to organize my thoughts on something I discuss with skeptical, curious researchers all the time.
This is my first attempt at articulating practical guidance for academics who want to use AI responsibly. I hope it’s useful, and I invite all feedback on how to make it better.