NeuralFlow

NeuralFlow : Workflow at the Speed of your Mind

** NLP Powered Process and Workflow Solution **

All of us have experienced process capturing frustrations at work - slow, missing info, and what the customer thinks they say and what you understand is very different.

For example - In my role as Amazon SDM, leading world class engineers and being responsible for processing billions of transactions per month, capturing processes and workflows has become 2nd nature.The end result is always the same - there's a process/diagram that is captured and entered in some system to execute on. But the initial experience is difficult, async, and not human-friendly.

What it does

NeuralFlow is meant to be used in the very early stages of process definition. It revolutionizes the experience in capturing processes and workflows we all need in everyday business life. NeuralFlow allows you to express in natural language the various bits and pieces of processes; when you have iterated, captured what you want - you can then save/export to other business enterprise tools.

How we built it

The tool is omni-channel. We architected an enterprise server where the user can experience NeuralFlow via various channels - directly on website, part of the ChatGPT Plugin Ecosystem (this demo), on mobile, or can be integrated via other tools that can consume the services.

We defined several agents in the backend that are governed by a Master Director agent - the NeuralFlow Director agent - responsible for the process/workflow generation. Rest of agents are LangChanin agents that can execute/integrate with any number of enterprise tools to generate appropriate transforms when exporting.

This makes this application infinitely scaleable for all verticals and cross cuts across all business needs where humans are involved to capture new processes/workflows.

Challenges we ran into

The ChatGPT plugin prompting and workflow (LOL) itself is different from regular prompting and regular software dev. OpenAI will act intelligently on your plugin API ends, it is astonishing.

Initially, building the Plugin itself was a bit tricky, but it is actually well architected.

We had problems deploying to a real server for now - we could not expose the openAI JSON file, we ran out of time for that and saving in the backend we had to pull back scope significantly so we don't run into problems.

Accomplishments that we're proud of

The overall flow is well done, and it demonstrates the heart of the problem/solution/value.

What we learned

The Plugin is actually an excellent way for customers to consume this kind of service.

What's next for NeuralFlow

Integration with other businesses in the backend. By Far - and anybody wants to throw us a sack of money and servers so we can actually run it on proper hardware?

Thank you.

GMATT * VALHERU on Discord. (Cosmin & Uros)

https://www.loom.com/share/49026d66540e4d31b759d56131f1a06f