ML × bioinformatics engineer — production AI for mRNA design and cfDNA cancer detection.
I'm Royce — an ML engineer working at the intersection of biology and AI. I ship end-to-end systems: codon-optimized mRNA design, cfDNA fragmentomics pipelines for ultra-early cancer detection, and production AI tooling (LLM backends, OCR, NL→SQL) that runs at scale.
- 🎓 M.S. Applied Data Science, University of Michigan · currently bioinformatics at CUHK
- 🧬 Day-job focus: mRNA design × AI (codon, UTR, structure, neoantigen, LNP) and cfDNA fragmentomics + methylation (0.001% VAF detection, multi-cancer)
- 🛠️ Also ship applied AI systems — production document AI (FastAPI, 12+ LLM backends), NL→SQL transpilers, agentic workflows
- 🌏 Hong Kong (HKT, UTC+8) · open to remote and relocation worldwide
- ML Engineer — applied AI / LLM systems
- Bioinformatics / Computational Biology Engineer — mRNA design, liquid biopsy, foundation models for biology
- Research Engineer — ML-for-biology, scRNA-seq, foundation models
- Part-time / consulting on cfDNA, mRNA tooling, formal methods
| Project | What it does | Why it matters |
|---|---|---|
| deepcatch | Performance-weighted multi-modal fusion for ultra-early cfDNA cancer detection (fragmentomics + methylation + sequence). | Targets 0.001% VAF sensitivity — multi-cancer screening before symptoms. |
| mrna-ai-toolkit | End-to-end mRNA design pipeline: codon optimization, neoantigen screening, scGPT embeddings, TrialGPT-style trial matching, LNP formulation. | Bridges foundation models with therapeutic mRNA workflows. |
| py-idp | Production AI document processing: 12+ LLM backends, self-hosted OCR, HITL review, Pydantic schemas, FastAPI. | Real-world applied AI — 92% coverage, fuzzers, public idp.console, AGPL-3.0. |
| deepcatch-methylation | CpG methylation-based cfDNA cancer detection — FinaleMe validation pipeline. | Companion to deepcatch, focused on the methylation channel. |
| speaksql | NL→SQL transpiler across 9 dialects (Postgres, MySQL, Snowflake, BigQuery, SQL Server, Databricks, DuckDB, SQLite, SAP HANA). | Engineering-heavy language tooling — write once, query anywhere. |
| ns-blowup | Lean 4 + Mathlib4 formalization toward Navier–Stokes regularity. | Differentiator: formal-methods chops you don't usually see on ML eng resumes. |
See all 54 public repos → for the full catalogue (cfDNA pipelines, quant signals, NL→SQL, formal math, LLM tooling).
Bioinformatics & computational biology
- cfdna-fragmentomics-pipeline — Tumor-naive cfDNA features (FSD, 4-mer motifs, DELFI/WPS, GC-correction); Cancer-vs-Healthy on FinaleDB.
- cancer-screening — ML / DL / RL for 0.001% VAF ctDNA detection.
- CMC — MICCAI 2024, robust semi-supervised multimodal medical image segmentation.
Applied AI & production systems
- nl2pbip — NL → Power BI report generator (v1.0.0, agentic self-reflection, Fabric integration).
- scrapex — AI-friendly web scraper (URL + schema → clean markdown/JSON).
- employee-voice-analytics — HR NLP toolkit (sentiment, emotion, theme, attrition risk).
Quantitative
- vcp-signals · AI-Trader · mr-strategy · vmaa — VCP patterns, mean-reversion, agent-native execution.
Production / ML: Python · PyTorch · JAX · FastAPI · Pydantic · Hugging Face · Docker · GitHub Actions Bioinformatics: scGPT · codon optimization · fragmentomics · methylation calling · LNP chemistry · single-cell pipelines Data: pandas · scikit-learn · Jupyter · R Formal / tooling: Lean 4 + Mathlib4 · LaTeX (Tectonic) · Bash · macOS
| 🦈 Achievements | Pull Shark · Quickdraw · Starstruck · Pair Extraordinaire |
| 🎓 Education | M.S. Applied Data Science — U-Michigan · bioinformatics — CUHK |
| 📦 Public repos | 54 · AGPL-3.0-or-later on the bigger ones |
| 🌍 Region | Hong Kong (HKT, UTC+8) · open to remote and relocation worldwide |
| 🏅 Open source | Production-grade code, fuzz-tested, semantic-versioned releases |
deepcatch — pooled ROC across cancer types (fig1 · repo)
deepcatch — sensitivity operating points (fig4) · mrna-ai-toolkit — codon → variant → neoantigen → trial → LNP → manufactured mRNA (SVG source)
py-idp HITL review (SVG) · speakSQL auto-discovered join graph (SVG)
Formal-methods hobbyist: Lean 4 + Mathlib4 — formal progress on Navier–Stokes regularity and Erdős #125 (sumset density). Copyleft by conviction — bigger repos ship under AGPL-3.0-or-later.
- 📧 roycelam@umich.edu
- 🆔 ORCID: 0009-0008-9113-769X
- 💬 GitHub Issues on any of the repos above — fastest way to start a thread
- 🌐 Open to remote worldwide · HK-based (UTC+8)
Last polished 2026-10. Edit history on the commits page.