LLMs and complex tasks
So I have building llm-flow-designer.com out of an own need. While building a product to do AppStore optimisation I was using a lot of complex, connected prompts to come up with a description and keywords matching current trends and the apps for which I was doing the optimization. I looked at many tools, frameworks and SDKs and while they were promising the world the reality underneath ended up in more work than before.
Instead of writing complex code following conditional execution graphs and loops I decided to make it visual. And not visual like dropping a few prompts and hope for the best but a more deterministic way of prompt and context engineering:
have a node for each action to take: LLM prompt execution, data pulling, conditional paths and loops
specify exactly the expected output (usually JSON) with fields and types
build quorums on field output by running it multiple times in parallel
implement conditions and loops to validate and run again
change and use different LLM models on the fly (test the differences)
have full traceability of time, token and cost usage
Then, next I started building a product for finance, extremely number centric (there is just true or false as valid answer) and it blew my mind.
This approach actually helped a lot here as well:
Changing prompts to cover more use cases, blow up execution graphs in code and any change in the chain could side effect heavily. The only way to master this was going visual with in-depth details available at your fingertips which made usually hours of debugging a 30 second problem!
Next is legal, data analytics and customer support for more concise and determined answered to difficult and complex questions fed by a gazillion systems but before jumping into that I ultimately decided to actually build my flow designer out as a whole own product on it's own and I’m looking for early access users harden what I have built for more use-cases: llm-flow-designer.com
So please drop a comment with your feedback and happy to reserve a front row seat!
