Posts about Architecture
16 posts tagged Architecture.
LLM security: prompt injection, data leakage, and what actually stops them
Most LLM security advice is 'write a better system prompt.' The actual defenses are architectural: what the model can see, what it can do, and what never returns to a user unchecked.
LLM application architecture: the layers that don't change with the model
Model, harness, data layer, and product surface: the four layers of a production LLM application, and which ones survive the next model upgrade.
Enterprise RAG: what changes with real documents, permissions, and scale
A RAG demo answers questions about a clean folder of PDFs. An enterprise deployment has to handle permissions per document, millions of pages, and daily churn.
RAG architecture: the pipeline from document to answer
Ingestion, chunking, embedding, retrieval, reranking, generation: the six stages of a production RAG pipeline, and where each one quietly determines answer quality.
What is RAG, in plain terms?
Retrieval-augmented generation, explained without the jargon: what problem it solves, the four steps in the pipeline, and where it stops being the right tool.
Enterprise AI agents: what changes at scale
The prototype agent and the enterprise one solve the same task with a different architecture underneath. SSO, audit, governance, and the review nobody skips.
Building AI agents that can safely modify customer data
Read access is forgiving of mistakes. Write access is not. The permission model, approval flow, and audit trail that make a write-capable agent safe to ship.
What are AI agents, actually?
Not a chatbot, and not magic: an AI agent is a model that can call tools and decide what to do next. What that definition includes, excludes, and implies.
How to architect an AI agent platform
Not one agent behind an API: a platform. The layers that stay stable while models and frameworks underneath them change, and the order to build them in.
Agent harness: what it is, and why your agent needs one
An AI agent is a model plus a harness. The model is rented; the harness is what you own, and it is where agents succeed or fail in production. What goes in one.
When a RAG system is the wrong answer
Retrieval-augmented generation is the default answer to adding AI to documents. The four situations where it is the wrong one, and what to build instead.
Context engineering is the new prompt engineering
Prompt wording matters less than what the model sees. Context engineering in practice: retrieval, summarisation, exclusion, and a budget for attention.
One agent is usually enough
Multi-agent systems are the fashionable architecture. Most problems need one agent with good tools; more is slower and harder to debug. When more is right.
The three architecture decisions that are expensive to reverse
Most early technical decisions can be changed later at reasonable cost. Three cannot. How to recognise them, and how to make them well in the first week.
Postgres is enough, until it is not
One database can be the queue, search index, vector store, and analytics layer for most products. The boring-stack case, and the signals that say add something.
How long does an MVP take? Six to ten weeks, if you cut
The honest answer to how long an MVP takes, what decides whether it is six weeks or six months, and the one activity in the first week that determines the rest.
Tell us what you are building.
We reply within one business day with how we would build it, what it would cost, and which engagement model fits.
- 01Tell us what you are building
A short form or an email. No deck required, and "not sure yet" is a fine answer.
- 02A call with an engineer
Within one business day. Technical questions get technical answers, from the person who would build it.
- 03A written scope and quote
Fixed price where the scope is defined. The document is yours whether or not you go ahead.