Diagramming Software Architecture: C4 vs. UML
Understand why modeling software matters and compare C4 and UML with practical examples for architects and developers.
WHAT WE BUILD / AGENT SYSTEMS
A modern reference architecture for operating enterprise AI agents.
Explore the path from a Python workflow to a governed, observable production agent. Select a step or a box to see what we design and build.
A PRACTICAL DELIVERY PATH · 6 STEPS
The architecture follows the job the agent needs to do. Choose the pieces that fit, then connect them behind clear operational boundaries.
REFERENCE ARCHITECTURE INTERACTIVE MAP
REQUEST PATH →
OBSERVE·EVALUATE·ITERATE
BUILD · DEPLOY
A Python agent, in a managed runtime
Pick a harness that fits the workflow, implement the agent in Python, then deploy it into an isolated runtime with a repeatable environment and session lifecycle.
Python · harness · isolated runtime
CONTROL · CONNECT
One governed path to tools
Connect the agent to the APIs and MCP tools it needs. Scope the catalog, broker credentials, and make each action allowed, approval-gated, or denied.
MCP · scoped tools · allow / ask / deny
GROUND · RETRIEVE
Relevant context from trusted data
Ingest approved documents, chunk and embed them, then retrieve only the context the agent needs. Keep source permissions and freshness in the design.
Ingestion · embeddings · vector search
ROUTE · PROTECT
Model access behind an AI gateway
Route requests to the right inference provider, centralize credentials, and apply budget, rate, privacy, and fallback controls around model traffic.
Provider routing · limits · cost visibility
OPERATE · IMPROVE
See each step in the agent run
Correlate the user request with retrieval, model, and tool calls. Track latency, failures, token use, and cost while respecting data-retention requirements.
OpenTelemetry · traces · logs · metrics
TEST · RELEASE
Check quality before it reaches users
Build a representative task set, score outputs against clear criteria, and compare versions. Release only when quality and operational limits are acceptable.
Task sets · regression checks · versioning
DESIGNED AROUND YOUR WORKFLOWPYTHON · MCP · OPEN STANDARDS
CODING CLOUD / INDEPENDENT NOTES
Software, AI & what we build with them.
Practical field notes on software engineering, AI, and building for the web.
LATEST ARTICLES · 7
Observability: Monitoring, Logging, and Tracing for Modern Systems
A comprehensive guide to implementing observability in software systems through monitoring, logging, and distributed tracing to understand system behavior and performance.
Database Architecture: OLTP vs OLAP Separation
Learn when separating transactional and analytical workloads can help scalable applications.
Proof of Concept (POC) Template: A Practical Guide
A comprehensive template for creating Proof of Concept (POC) documents, focusing on hypothesis-driven experiments with clear success criteria and time-boxed execution.
Software Requirements Specification (SRS) for Web Applications: A Practical Template
A practical web application requirements template, with functional and non-functional requirements, interfaces, constraints, and architecture diagrams.
Software Project Handover
A practical guide to executing a smooth and professional software project handover.
EU AI Act: A Practical Guide for AI-Powered Web Applications
A scannable, actionable guide to EU AI Act compliance for web applications using AI systems. Features a comparison table for Generic Business, FinTech, and HealthTech apps.