Computational Biology / Platform Engineering
Scientific systems that survive production.
Current Focus
Biology, ML, and infrastructure in the same room.
The work sits between scientific reasoning and production reliability: molecular modelling, NLP, data contracts, APIs, relational schemas, and operational pipelines.
Featured Projects
Selected work
DASH
Internal LeverageHR
Partner PlatformCompatible across the spectrum
Scoring Copilot
Proposes a mechanism. Then tries to prove itself wrong.
A real run of the research pipeline that discovers and grades candidate scoring features — one agent proposes a mechanism-of-action hypothesis, a second tries to falsify it, and the survivors are graded across five evidence axes before they reach production scoring.
scoring-copilot · feature discovery
sql-executor
Named, parameterized, pre-reviewed SQL queries for common lookups, plus a raw-SQL escape hatch for one-off schema exploration.
translate-ids-mcp
Cross-translates identifiers (kit-id, user-id, test-id, external-id) across internal databases, optionally joined to analysis-id.
expression-mcp
Pulls filtered expression data (by feature type and read-count threshold) from the bioinformatics database as a clean, ready-to-analyze CSV.
cohort-exploration-mcp
Queries Viome's internal biological and customer metadata to check whether a hypothesized relationship actually shows up in real cohort and label data.
kegg-mcp
Looks up pathway and mechanism-of-action data from KEGG — the biological grounding for a hypothesis, not just a citation.
literature-review-mcp
Full-text literature retrieval with custom embeddings and ranking — built because abstract-only search missed too much mechanism-level nuance.
Research Agent
Proposes a mechanism-of-action hypothesis for a candidate feature — why it should plausibly relate to the target outcome, biologically.
Critique Agent
Actively tries to disprove the hypothesis: searches the literature for contradicting evidence and cross-checks it against internal cohort and label data.
Evidence Tiering
Grades surviving hypotheses across five axes: internal correlation strength, mechanism plausibility, actionability, strength of evidence, and druggability.
Explore/Exploit Tuning
Balances the research/critique loop so it surfaces novel hypotheses without wasting cycles re-litigating obvious dead ends.
GitHub Activity
Past year
767 Contributions
72 Pull Requests
40 Repos
Language Mix
- Python43.0%
- Java38.3%
- HCL9.4%
- Shell2.8%
- R2.0%
- Go1.6%
- Other2.9%
What I Build
Production-grade scientific platforms.
Flask APIs, relational databases, scalable curation scripts, NLP models, multiomics analysis workflows, and molecular simulation tooling with clear handoff boundaries.
Looking For
Senior data science or platform roles.
Best fit: biotech, health-tech, AI drug discovery, precision medicine, or infrastructure teams building for scientific workloads.