NeedGen
A multi-agent pipeline that turns raw product ideas into structured requirements.
- 4
- Cooperating agents
- $0.007
- Cost per full run
- 10+
- Product concepts tested
NeedGen is a GenAI pipeline that takes a half-formed product idea and walks it through structured reasoning until it comes out the other side as something an engineering team can actually plan against. The interesting constraint wasn't 'can an LLM write requirements' — it was making the output reliable, decomposable, and cheap enough to run on every idea, not just the precious ones.
The problem
Product requirements are where good ideas quietly die. They're either too vague to build or written so late that the team has already guessed. I wanted a system that meets an idea at its messiest and produces a clear, structured spec — fast and cheap enough to be disposable.
How it works
- A four-agent pipeline, each agent owning one stage: idea analysis, feature decomposition, specification generation, and review.
- Built on LangChain + CrewAI so agents hand off structured context rather than re-deriving it, keeping each call tight and cheap.
- Designed developer feedback loops so teams can refine generated requirements and align outputs with real engineering planning.
What I learned
Cost discipline is a product feature. Running at ~$0.007 per execution meant the team could treat requirement generation as something to do constantly and casually — which changed how it got used. Cheap and good beats expensive and perfect.