Allan — Technomancer of Systems, Science, and Code

I build reproducible ML infrastructure for scientific intelligence.

Independent ML engineer and systems architect focused on autonomous research tooling, distillation pipelines, and HPC-ready inference stacks. Everything I ship is designed for disposable compute, durable results, and transparent provenance.

Current focus: evolving NexaCompute and scaling Nexa_Sci_distilled_Falcon-10B into a production-grade scientific assistant.

NexaCompute

NexaCompute Infrastructure

Disposable compute. Durable lineage. NexaCompute orchestrates data prep, distillation, distributed training, LLM-as-judge evaluation, dashboards, and deployment across any GPU provider. The platform remains available for reference but is no longer actively maintained.

  • End-to-end manifests for distillation, training, evaluation, and serving.
  • Modular stacks covering nexa_data, nexa_distill, nexa_train, nexa_eval, nexa_ui, and nexa_infra.
  • Provider-agnostic orchestration patterns for Lambda, CoreWeave, RunPod, AWS, and bespoke clusters.
  • Historical telemetry and cost tracking for reproducible scientific ML runs.
GitHub Repository

Skills

Programming Languages

  • Python
  • Julia
  • C++
  • Rust

Machine Learning Frameworks

  • TensorFlow
  • PyTorch
  • JAX
  • Scikit-learn

Quantum Computing

  • Qiskit
  • Cirq
  • Pennylane

Scientific Computing

  • NumPy
  • SciPy
  • SymPy
  • Matplotlib