Architecture
19 posts tagged Architecture — engineering notes and field reports from the TechKis team.
Posts in Architecture
19 posts
- 10 min read·AI EngineeringArchitectureSecurity
Self-hosted LLMs vs provider APIs — when data residency actually forces your hand
Self-hosting an open-weight model is usually the wrong default and occasionally the only option. Here's what genuinely forces it — regulation, contracts, volume economics — and the hybrid split most teams land on.
Read post - 9 min read·AI EngineeringArchitectureAPIs
MCP vs A2A vs HTTP APIs — what's actually different, and when each one wins
MCP, A2A and plain HTTP APIs solve overlapping but distinct problems. Here's what each protocol is really for, how they compose, and how to pick without cargo-culting the newest acronym.
Read post - 9 min read·AI EngineeringProductSaaS
Why every SaaS will have AI agents by 2027 — and what that changes about how you build
Agents are moving from bolt-on chat widgets to the primary way users get work done inside software. Here's the shift that's driving it, what an agent-native SaaS actually looks like, and what it means for your architecture and moat.
Read post - 9 min read·AI EngineeringArchitectureDeveloper Experience
Building AI features without LangChain — the raw-SDK path most teams should take first
You don't need a framework to ship an AI feature. Here's what the raw provider SDK actually gives you, the small amount of glue you write yourself, and when a framework finally earns its place.
Read post - 10 min read·AI EngineeringRAGRetrieval
RAG is not dead — you're just building it wrong
"Just use a bigger context window" doesn't kill RAG. Bad chunking, no reranking, and retrieval you never evaluate kill RAG. Here's what a retrieval pipeline that actually works looks like.
Read post - 9 min read·AI EngineeringMemoryArchitecture
AI memory explained — short-term vs long-term, and why the difference matters
"Memory" in AI apps means two very different mechanisms with different costs and failure modes. Here's what short-term (context) and long-term (retrieval/state) memory actually are, and how to design both.
Read post - 10 min read·AI EngineeringArchitectureBackend
How AI changes backend architecture — the parts that are genuinely different
Adding an LLM to your backend breaks assumptions that held for a decade: latency, determinism, cost per request, statefulness. Here's what actually changes and how to design for it.
Read post - 10 min read·AI EngineeringArchitectureEnterprise
The AI gateway pattern for enterprise applications
Letting every service call model providers directly is how enterprises lose control of cost, security and compliance. An AI gateway is the single seam where routing, auth, budgets, logging and guardrails live. Here's how to build one.
Read post - 10 min read·AI EngineeringObservabilityArchitecture
AI observability — logs, traces, cost and hallucinations
You can't run an LLM system on hope. Traditional observability misses the things that actually go wrong with AI: silent quality drift, runaway token cost, and confident wrong answers. Here's the observability stack an AI system needs.
Read post - 9 min read·AI EngineeringCostArchitecture
AI rate limiting and cost control — before the bill surprises you
One buggy loop or abusive user can turn an LLM feature into a five-figure invoice overnight. Rate limiting and cost control for AI aren't the same as for a normal API — here's how to cap spend without breaking the product.
Read post - 11 min read·ArchitectureScalingBackend
Scaling from 10 users to 10 million — the decisions that matter at each stage
Scaling isn't one problem. It's a sequence of different problems that appear at different stages. Here's what actually matters at 10 users, 10,000 users, 100,000 users, and 10 million — and the mistakes that come from solving tomorrow's problem today.
Read post - 8 min read·API DesignArchitectureBackend
API-First Development — why the contract should come before the code
API-First means designing the contract before writing the implementation. Here's why that order matters, what it changes about how teams work, and the practical workflow that makes it work without slowing you down.
Read post - 9 min read·ArchitectureDDDStartups
Domain-Driven Design for startups — the parts worth keeping, the parts worth skipping
DDD is a powerful set of ideas buried under a lot of enterprise ceremony. Here's what's genuinely useful for a startup — bounded contexts, ubiquitous language, aggregates — and what you can safely skip until you need it.
Read post - 8 min read·ArchitectureHexagonal ArchitectureSoftware Design
Hexagonal Architecture without overengineering — ports, adapters and the line you shouldn't cross
Hexagonal Architecture is one of the most useful ideas in software design. It's also one of the most over-applied. Here's what the ports-and-adapters model actually buys you, where it becomes ceremony, and how to apply it without drowning in abstractions.
Read post - 9 min read·ArchitectureClean ArchitectureBackend
Clean Architecture in real projects — what survives contact with a deadline
Clean Architecture looks elegant on a diagram. In production it gets messy. Here's what the layers actually buy you, where the theory breaks down, and the pragmatic version we apply on real client work.
Read post - 10 min read·ArchitectureMicroservicesBackend
Monolith vs Modular Monolith vs Microservices — picking the right shape for your stage
A practical breakdown of when a monolith is the right call, when to modularise it, and when microservices actually earn their operational cost — with the signals that tell you it's time to move.
Read post - 9 min read·LangGraphLangChainOpenAI
LangGraph vs LangChain vs raw OpenAI SDK in 2026 — what we actually pick
An opinionated, code-first comparison of LangGraph, LangChain and the raw OpenAI / Anthropic SDKs for shipping production agents in 2026 — when each one earns its keep, where they bite, and what we default to on new client work.
Read post - 9 min read·DesktopTauriElectron
Tauri vs Electron in 2026 — binary size, RAM, auto-update and the parts no one benchmarks
A practical comparison of Tauri and Electron for desktop apps in 2026 — measured binary size and RAM, the real auto-update stories, code-signing pain, and where each one earns its keep.
Read post - 9 min read·AWSBedrockLLM Infrastructure
AWS Bedrock vs direct provider APIs — cost, lock-in and the procurement reality
When to call OpenAI and Anthropic directly versus going through AWS Bedrock — a practical breakdown of cost, latency, model availability, governance and the procurement reality that actually decides the answer on most enterprise projects.
Read post