Board Ready
Led two interns analyzing 500+ SEC DEF 14A filings with BERT classifiers and LLaMA-2 70B to surface board-diversity data across the S&P 500.
- ML
- NLP
- BERT
- LLaMA-2
- Python
- SEC/EDGAR
Board diversity data for the S&P 500 exists — it’s just buried in prose, inside DEF 14A proxy statements, in a different format for every company. I led a team of two other interns extracting it at scale.
The problem
A proxy statement is a long legal document written for humans. Board composition might appear as a table, a paragraph, a set of director biographies, or all three, with no consistent schema across filers. Pulling 500+ of them from SEC EDGAR and getting comparable structured data out is not a parsing problem; it’s a reading-comprehension problem.
The approach
We used two model families for two different jobs:
- BERT classifiers for the high-volume, well-defined decisions — is this section about board composition, is this a director biography — where a fine-tuned encoder is fast, cheap, and consistent across hundreds of documents.
- LLaMA-2 70B for the passages that needed actual comprehension, where the information is stated once, in prose, in a way a classifier can’t reach.
Routing the easy cases to the small model and reserving the large one for the hard ones is the difference between a pipeline that finishes and one that doesn’t.
Leading the team
This was my first time leading engineers rather than being one. The lesson that stuck: with two people and a fixed number of weeks, the cost of an unclear interface between their work is much higher than the cost of spending a morning defining it. Splitting the pipeline into stages with agreed-upon shapes between them let all three of us work at once.