About
I work at the intersection of machine learning and protein engineering, building computational pipelines that help get better therapeutic proteins into wet-lab validation faster.
My current focus is at BioLM, where I develop generation and scoring workflows for protein variant design — multi-method generation strategies, scoring ensemble design, in silico validation, and degradation risk assessment. The technical reports on this site come directly from that work.
My stack: transformer architectures, LoRA/PEFT fine-tuning, large biological dataset curation, scalable model serving (Kubernetes/GKE), and the BioLM API ecosystem. I care about pipelines that are actually reproducible — DuckDB caching, resumable runs, real numbers rather than directional guesses.
Publications & Preprints
-
Towards a Dataset for State of the Art Protein Toxin Classification
— CA Challacombe, NS Haas. bioRxiv, 2024.
Transformer-based sequence models for classifying protein toxins from sequence alone.
Selected work
- Adaptyv Bio EGFR Binder Competition — Computational design of EGFR binders using multi-method generation and ensemble scoring; designs submitted for synthesis and experimental validation.
- Bio × ML Hackathon (Evolved2024) — Team Silica.
Consulting
I take on consulting work in protein design pipelines, ML model evaluation, and generation strategy. If you're building in the protein engineering or biotech ML space and want an experienced hand, reach out.