Reproducible platforms that stay calm in production.

Problem

Agronomists and data scientists needed reliable, high-throughput access to geospatial agricultural models and environmental data at continental scale — with strict SLOs, replay-safe data paths, and zero tolerance for silent failures in decision-critical pipelines.

Approach

Designed and delivered FastAPI microservices for agricultural data and model-execution workflows. Owned distributed architecture on AWS (ECS/EKS, DynamoDB, SQS/SNS, SSM, ECR, IAM). Built SQS-based async ingestion with batching, DLQs, replay safety, and checkpointing. Implemented federated GraphQL platform (Apollo Federation) with schema governance across subgraphs. Integrated BigQuery analytics telemetry with cost-optimized query patterns. Shipped Prometheus/Grafana/Loki/OpenTelemetry SLO dashboards and incident runbooks.

Impact

Promoted to Senior Software Engineer (Nov 2024). Services power model execution and geospatial data delivery across Bayer's North American precision agriculture operations, processing field data at continental scale across millions of acres annually. The SQS async ingestion layer handles high-concurrency model runs with fault-isolated DLQs and replay-safe checkpointing, keeping error budgets tight under variable seasonal demand. Federated GraphQL platform unified multiple engineering teams under shared schema governance, reducing cross-team API integration friction. Service templates and internal SDKs adopted org-wide shortened new service bootstrap time from days to hours. Observability uplift with Prometheus/Grafana/Loki cut mean time to detection for production incidents and gave on-call engineers a direct path from alert to runbook.

Problem

Computational cell biologists needed a unified platform to model and simulate biochemical networks, reaction-diffusion equations, stochastic processes, and electrophysiology — across deterministic ODE, PDE, and particle-based simulation regimes — without managing heterogeneous solver environments manually.

Approach

Contributed to the VCell core platform (Java, 113+ stars, 13K+ commits) and its solver suite (vcell-solvers: Chombo adaptive mesh, NFsim network-free stochastic, Smoldyn molecular dynamics, Hy3S hybrid stochastic, finite-volume PDE solvers). Built SLURM-based job dispatch infrastructure for HPC cluster simulation execution — handling submission, monitoring, lifecycle management, recovery, and checkpointing. Containerized solver environments with Docker and Singularity for reproducible cross-platform execution. Published the SLURM dispatch client to PyPI.

Impact

VCell is actively used by cell biology research groups at universities and institutes globally, with simulations backing peer-reviewed publications across calcium signaling, cytoskeletal dynamics, and membrane transport. SLURM dispatch infrastructure brought reliable job lifecycle management to HPC-backed simulations — concurrent submissions across multiple solver types, automated recovery from transient cluster failures, and checkpointed state that prevents lost compute on long-running stochastic and PDE runs. Containerized solver environments reduced cross-platform setup overhead and made new solver types reproducibly deployable without manual environment configuration. Infrastructure work directly enabled RunBioSimulations and BioSimulators — both published in Nucleic Acids Research with 73 combined citations.

Problem

Computational biologists needed a single web platform to share, reuse, and run biomodeling studies across dozens of heterogeneous simulation engines — SBML, CellML, NeuroML, SED-ML, and more — without installing each tool locally or managing HPC job submission manually.

Approach

Built on top of VCell's solver infrastructure: designed the SLURM dispatch server (submission, monitoring, lifecycle, recovery, checkpointing) that connects the web platform to HPC cluster simulation engines. Developed Flask and NestJS APIs for simulation setup, metadata persistence, and results retrieval. Containerized simulation environments with Docker and Singularity. Published the SLURM dispatch client to PyPI. Integrated NATS messaging for distributed coordination.

Impact

Published in Nucleic Acids Research 2021 (IF ~14). 73 total citations across the BioSimulations/BioSimulators ecosystem. Platform (biosimulations.org) supports sharing and reproducibility of entire biomodeling studies across continuous, discrete, stochastic, and logical model types. 42+ stars on GitHub.

Problem

No curated, machine-readable registry of standardized, containerized biosimulation tools existed. Tool versions and command-line interfaces were undocumented and incompatible — reproducibility suffered at community scale when researchers tried to reuse each other's models.

Approach

Contributed to the registry schema, validation pipeline, and API surface at biosimulators.org. Each simulator entry includes version provenance, standardized command-line and Python interfaces, and validated Docker container images. Built a test suite (Biosimulators_test_suite) and utilities (Biosimulators_utils) to validate that simulation tools implement BioSimulators standards before registration.

Impact

Published in Nucleic Acids Research 2022. Established as the canonical registry for simulation engine discovery in the computational biology community — covering SBML, CellML, NeuroML, SED-ML, and OMEX formats. Cited alongside BioSimulations in the shared 73-citation body of work. 14+ stars on the core registry repo.