I’m interested in what happens after the model works.

CairnOps

Agentic expedition planning

Turns a free-text expedition request into route, weather, equipment and safety decisions through six domain agents.

Engineering focus Conditional LangGraph routing, Qdrant retrieval, LiteLLM fallback, evaluation and end-to-end tracing on a production-like GKE setup built under zero-cost constraints.

• Runtime image cut from 4.7 GB to 121 MB via multi-stage builds
• Eval gate: fact coverage ≥ 0.60, risk-level accuracy ≥ 0.75 (Ragas + DeepEval)
• Six-agent LangGraph pipeline with conditional risk routing

OrbitDecay

Satellite decay risk platform

Forecasts satellite decay risk through a checkpointed MLOps workflow spanning fifteen services.

Engineering focus Moving CPU-heavy orbital calculations from Python into a Rust and Polars acceleration layer, then serving physics-informed forecasting through FastAPI.

• Eight-phase checkpointed pipeline orchestrating 15 microservices
• Rust + Polars acceleration layer for CPU-bound orbital calculations
• Physics-informed LSTM forecasting served through FastAPI

LinguaGate

AI outreach workflow

A two-day prototype for lead intake, enrichment, scoring, routing, personalized outreach and CRM tracking.

Engineering focus Keeping a fast prototype inspectable with typed state, clear agent boundaries, MLflow experiments, Langfuse traces and a small FastAPI surface.