Posts about Engineering
12 posts tagged Engineering.
LLM observability: what to log, and what it actually catches
Traditional APM tells you a request was slow. LLM observability has to answer a different question: was the answer right, and why did the model say that?
How to build an AI analytics assistant
'Ask your data a question' is a text-to-SQL problem wearing a chat interface. The architecture that makes it trustworthy: schema grounding, guardrails, and a query the user can check.
How we reduced LLM costs in a multi-tenant AI platform
One customer's heavy usage was inflating everyone's bill. The per-tenant budgeting, routing, and caching changes that cut the platform's running cost by 58%.
LangGraph vs CrewAI vs AutoGen: a practical comparison
Three popular agent frameworks, compared on what actually differs in production: control over the loop, debugging, and how much they decide for you.
MCP servers: exposing your systems to models safely
The Model Context Protocol is now how tools reach a model. How we design an MCP server for a client system: scope, safety, and what to leave out at first.
AI coding agents in production: the review gates that make them safe
Coding agents now write a lot of software. What we put between an agent's output and production: five review gates, why each exists, and what we measure.
What we check before taking over someone else's codebase
The audit we run in the first week on an inherited codebase: the eight questions we answer, in order, and what each one tells you about the months ahead.
Small models and routing: how we keep AI features affordable
The demo used the biggest model for everything. Production cannot. How routing, caching, and small tuned models cut the running cost of AI features by half.
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.
What breaks in AI-generated codebases, in the order it breaks
We have audited many codebases built mostly by AI tools. The same seven problems appear in the same order, from week one to month six. How to catch them early.
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 to hand over a codebase so a team can run it
Ownership is a clause; handover is the work that makes it real. The checklist we run at the end of every engagement so your team can run the system without us.
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.