Decisions, Not Models
The architectural decisions that separate production-grade engineering from experimental code.
The Economics of Intelligence: FinOps for AI and the Real Cost of a Single Prediction
Why your model isn't expensive — your decisions are. Inference cost optimization, model compression, and cost-per-decision as a first-class engineering SLO.
Decisions, Not Models · Issue #14The Model Can No Longer Refuse to Explain Itself: Governance, Explainability, and Compliance-as-Code in Production MLOps
Automation without provenance is faster risk. Model registries, generated model cards, policy-as-code deployment gates, explanation drift and the EU AI Act.
Decisions, Not Models · Issue #13Combating Model Decay: Data Drift and the Architecture of Continuous Training
ML systems fail silently, returning 200 OK while rotting. Data drift, concept drift and the architecture of Continuous Training with champion-challenger gates.
Decisions, Not Models · Issue #12Beyond the Single Node: Orchestration, Autoscaling, and the Distributed AI Architecture
One container hits a ceiling. Kubernetes orchestration, autoscaling, scale-to-zero economics, shadow deployments and canary releases for machine learning.
Decisions, Not Models · Issue #11Beyond the Artifact: Containerization and High-Performance Model Serving
Versioning the model is not versioning its world. Docker for environment parity, immutable infrastructure and FastAPI serving with Pydantic data contracts.
Decisions, Not Models · Issue #10The Pillars of MLOps: Bridging the Gap Between Code, Data, and Models
Git alone cannot version machine learning. Code, data and models as a volatile triad — DVC, MLflow and the architecture of true reproducibility.
Decisions, Not Models · Issue #9Beyond the Notebook: The Imperative of MLOps in Modern AI Architecture
A model in a notebook has zero business value. Why MLOps exists, how it differs from DevOps, and the hidden technical debt that silently erodes ML systems.
Decisions, Not Models · Issue #8The Ghost in the Machine: Navigating Graph Observability and Relational Explainability on GCP
Autonomous agents need audits. Minimal sufficient subgraphs, counterfactual analysis and SubgraphX for relational explainability on the GCP stack.
Decisions, Not Models · Issue #7The Closed Loop: How Autonomous Agents Dynamically Rebuild Graph Realities
From read-only RAG to read-write agency: autonomous agents that mutate graph topology through Cloud Run tools, Eventarc feedback and Vertex AI reasoning.
Decisions, Not Models · Issue #6The Signal and the Noise: Mastering Graph Attention Networks for Enterprise Decision Systems on GCP
Not every edge carries signal. How Graph Attention Networks learn to weight relationships, with the GAT attention equation and TPU-backed serving on GCP.
Decisions, Not Models · Issue #5Beyond Distance: Engineering Similarity and Clustering in Non-Euclidean Graph Systems on GCP
Distance stops meaning similarity in graphs. Structural embeddings, cosine similarity and BigQuery ML clustering for non-Euclidean segmentation on GCP.
Decisions, Not Models · Issue #4The Arrow of Time in Networks: Conquering Temporal Data Slippage with Feature Stores
Graph topologies change faster than models perceive. Point-in-time correctness and Vertex AI Feature Store as the fix for temporal data slippage in GNNs.
Decisions, Not Models · Issue #3The Serverless Equation: Conquering the Cold Start in Real-Time AI Inference
Cold starts cripple real-time AI inference. Engineering serverless GNN deployment on GCP with Cloud Run, minimum instances and Eventarc orchestration.
Decisions, Not Models · Issue #2Beyond the Vector: Why Graph Neural Networks are the Strategic Choice for Enterprise Generative AI on GCP
Why GNNs beat flat vector search for enterprise GenAI: GraphRAG for multi-hop reasoning, plus MLOps on GCP with BigQuery Graph, TPU acceleration and Vertex AI.
Decisions, Not Models · Issue #1