Fine-Tuning vs RAG vs Prompting: A Decision Framework for Cost and Quality
Choose prompting, retrieval, fine-tuning, or a combination by separating knowledge from behavior—and measuring their full cost and quality trade-offs.
Architecture, cloud & AI
Architecture write-ups with the HLDs, LLDs, sequence diagrams and table designs left in — on cloud, AI agents, LLMs and the systems around them.
Choose prompting, retrieval, fine-tuning, or a combination by separating knowledge from behavior—and measuring their full cost and quality trade-offs.
01 — Topics
Kubernetes, Docker, microservices and modern architecture patterns — HLDs, LLDs and the trade-offs behind them.
36 articles · 7.4 h
EC2, Lambda, ECS, CDK, Well-Architected reviews and cost optimisation in practice.
5 articles · 1.4 h
LangGraph, CrewAI, AutoGen and agentic design patterns that survive production.
7 articles · 1.9 h
RAG, fine-tuning, evaluation and prompt engineering for real systems.
5 articles · 1.4 h
Model Context Protocol servers, tool integrations and secure agent plumbing.
4 articles · 1.1 h
YouTube-backed walkthroughs with code, diagrams and demos.
0 articles · 0 min
Architecture styles from monolith to cell-based, sized small to enterprise, with HLDs, LLDs and trade-offs.
17 articles · 3.6 h
Cloud Run, GKE, Spanner, BigQuery and Vertex AI blueprints, from a single service to an enterprise foundation.
5 articles · 1.3 h
App Service, Container Apps, AKS, Cosmos DB and Fabric architectures with Bicep and landing zones.
5 articles · 1.4 h
Warehouses, lakehouses, ELT/ETL, CDC and streaming pipelines with modelling and data quality.
5 articles · 1.4 h
Semantic layers, dashboards, product analytics and governed self-serve BI.
5 articles · 1.4 h
OpenTelemetry, logs, metrics, traces, SLOs and incident response at every scale.
4 articles · 1.2 h
Requirements to release: testing, CI/CD, IaC, GitOps, platform engineering and supply-chain security.
8 articles · 2.0 h
02 — Latest
A practical design for testing agent outcomes and tool-use trajectories offline, then turning the results into trustworthy release gates.
Practice 12 system design interview questions on tenant isolation, fair resource sharing, and the trade-offs behind multi-tenant platforms.
A practical testing strategy for a medium-scale SaaS: use unit, integration, contract, and end-to-end tests at the right points in the SDLC, with Testcontainers for realistic dependencies.
Prepare for interviews on software supply-chain security, from dependency pinning and SBOMs to signed provenance, deployment policy, and pipeline observability.
A practical AWS EKS blueprint for running Loki, Grafana, Tempo, and Mimir with bounded retention, controlled cardinality, safe releases, and useful SLOs.
A medium-scale AWS blueprint for governed product events, RudderStack routing, replayable S3 data, and analytics models in Redshift.
Thirteen interview questions on governing MCP capabilities across hosts, servers, identities, policy enforcement, audits, and production operations.
A practical design for a medium-scale retail warehouse that preserves history with SCD Type 2 and keeps sales, inventory, and returns consistent through conformed dimensions.
A 12-question study guide to building reliable LLM contexts, resisting prompt injection, and keeping model-generated outputs inside application trust boundaries.
A medium-scale Azure design for accepting documents, orchestrating durable processing, persisting results, and operating the workflow safely across the SDLC.
Twelve practical Google Cloud interview questions on Pub/Sub semantics, failure handling, replay, capacity, and designing event-driven systems that recover safely.
A medium-scale GCP SaaS architecture that combines Cloud Run, Cloud SQL, Pub/Sub, Memorystore, and Cloud Tasks across the software delivery lifecycle.
Newsletter
The week’s new long reads on architecture, cloud and AI — with the diagrams left in. Delivered Saturday morning, no noise in between.