Sign in Get started

Sign in Get started

Docs

Requirements Engine

From knowledge base and documents to functional specs and ClickUp-ready backlogs — with humans reviewing every artifact.

The Requirements Engine converts product knowledge — knowledge-base articles, Markdown, HTML, PDFs, DOCX, URLs — into a functional specification and a structured backlog.

Pipeline

Ingest sources into a canonical KnowledgeArticle model.
Extract requirement candidates, capabilities, workflows, actors, and rules.
Build a product map / knowledge graph and a FunctionalSpecification.
Plan epics, stories, and tasks with traceability.
Export Markdown/backlog artifacts ready for ClickUp (via MCP or review).

Why it is powerful

It compresses weeks of analysis into a repeatable pipeline — without pretending the model “understands your company” by magic. You choose the source, the strictness mode, and which artifacts become work.

You stay the pilot

CLI flags control ingest-only, strict validation, and filters.
Validators surface inconsistencies instead of silently inventing scope.
LLM gateways (Ollama, Claude, OpenAI, stub) are swappable tools behind ports.

Stack highlights

TypeScript, hexagonal architecture, Vitest, AJV, Cheerio/Crawlee for ingestion, multi-provider LLM adapters.

A message from Xavi:

founder of BeSmart in 2014

I have been working in artificial intelligence since 2012.

Over these years I have watched several waves of enthusiasm come and go. Technologies that looked revolutionary, promises that everything would change overnight — and then, reality. Today we are living through the biggest AI boom yet, and it is genuinely impressive. But it is also creating expectations that do not always match reality.

Every week I speak with companies that have invested time and money in AI projects that did not deliver. Many come to us after a bad experience. Not because artificial intelligence does not work, but because someone made them believe that adding "AI" to a process was enough to make everything better.

It does not work that way.

Artificial intelligence is not a magic product. It is an extraordinarily powerful tool — but it is still a tool. And like any tool, it only delivers good results when applied to well-understood processes, with clear goals and a solid methodology.

The curious part is that our process is not revolutionary at all.

In fact, I have to admit we have not invented anything.

We apply principles that have worked for more than twenty years in engineering, digital transformation, and continuous improvement. The only thing that has changed is technology’s ability to automate tasks that were previously impossible.

Our work starts by understanding how a company actually works. We analyze processes, identify bottlenecks, map the technology landscape, find automation opportunities, implement improvements incrementally, measure results, and put governance in place so the system keeps evolving in a controlled way.

That order matters.

Because a disorganized company with artificial intelligence is still a disorganized company. It just makes the wrong decisions much faster.

Our goal has never been to sell artificial intelligence.

Our goal is to help companies work better, make better decisions, and build an organization ready for the future. If AI is the best way to get there, we will use it. If it is not, we will say so.

We believe the real value is not having the most advanced AI model, but knowing where, when, and why to use it.

That is what we do at BeSmart.