AI Approach Finder
Not every problem needs an LLM. Answer eight questions and get a scored recommendation across two layers: the right AI brain and the right architecture to deploy it.
About the AI Approach Finder
Who it is for
Use this tool when a team has a problem to solve and the conversation has jumped straight to a technology. It is for architects, product managers and engineering leads who need a reasoned starting point before choosing a vendor or a framework.
It takes about three minutes and gives a recommendation you can challenge, which is more useful than a preference nobody has examined.
What it measures
Eight questions about the task, your data, autonomy, explainability, latency, input complexity and team constraints produce a scored recommendation on two layers.
- The reasoning layer. Business rules, machine learning, deep learning or generative AI. This is the kind of intelligence the problem needs.
- The architecture layer. Direct integration, retrieval-augmented generation (RAG) or agents. This is how that intelligence reaches your data and takes action.
How to read the result
Every answer adds weight to the options it favours. The result shows a score for each option on both layers, so you see the runner-up as well as the winner.
A clear winner means your constraints point one way. Two close scores mean the choice depends on something the questions could not capture, usually cost or the skills you already have. That is the discussion to have with your team.
Take a recommendation of business rules seriously. If the logic is fixed and explainability is critical, a deterministic solution is cheaper, faster and easier to audit than a model.
Common questions
When should I use RAG instead of agents?
Use RAG when the system needs to answer from your private documents or knowledge base and only has to respond. Use agents when it has to plan and carry out several steps with tools. Agents add cost, latency and failure modes, so choose them only when the task needs that autonomy.
Do I always need generative AI?
No. Predicting a number or classifying structured records is a machine learning problem. Applying fixed validation logic is a rules problem. Generative AI earns its complexity when the input or output is natural language.
What if I have no labelled data?
That rules out training your own model for now and points to rules or to a pre-trained generative model. The tool accounts for this in its second question.
Is it free?
Yes. The tool is free and runs in your browser with no account. The downloadable report asks for an email address, and your scores are sent with it so I can follow up if you want to talk them through.