I take on contract and consulting work: recommender and ranking
design, data science on large or messy sources, measurement, and applied work with
language models.
- Recommender and ranking.
- Designing one, or auditing one you already run: what its objective selects for.
- Measurement.
- Construct definition, codebook development, human coding, and inter-rater agreement.
- LLM labelling at scale.
- Classification and annotation across corpora too large to read, validated against human coders on a stratified sample.
- Data engineering.
- Multilingual, malformed, and very large sources turned into something a team can query, entity resolution included.
- Knowledge graphs.
- Ontology-driven extraction from text with entity normalisation and coverage checking, benchmarked against a public dataset.
- Matching and allocation.
- Assignment and allocation problems solved to optimality, with a review interface so a person can overrule the result.
- Interpretability.
- Mechanistic work on how a model arrives at its output. For when test-set performance isn't enough to justify a decision.