💭 What Am I Working On?
Lately I've been writing fewer posts — the mood isn't quite right.
Wars, conflicts… They don't add optimism.
But life goes on, and that seems to be the point. Keep going. Hug your family. Do what needs to be done.
So here's what I've been up to. I'm building a service that generates business plans. Sounds simple — especially in the age of AI chats. Write a prompt, get a plan. Why would anyone need anything else?
I think I know. Let me share a bit of how my service actually works. A few principles that ensure the output is a document you can take to a bank.
– A baseline expert document skeleton that is completely independent of any AI – User inputs are processed to identify a business archetype from a large data dictionary, determining the properties and parameters of the business – To determine specific parameter values — rent costs for a given location, labor fund based on required skills, patent costs, current income tax rates, etc. — the service queries a dedicated data search agent – The search agent returns an array of data and sources to the main service, forming the core context. Key point: AI is not involved in data collection at all – Data processing: financial and mathematical modeling happens inside Python scripts; market and competitor analysis is performed by an AI model running at temperature ≤ 0.2 (which ensures precise interpretation of numbers and eliminates data manipulation) – Document section generation passes through a strictly defined flow, with a verification step checking for digital noise and AI artifacts
The core idea: isolate the AI from forming the context. And I do this at the algorithm level. Only once the context has been built from a concept framework, archetype dictionaries, and live search data — does the model process it and generate text.
The technology is called RAG — Retrieval-Augmented Generation.
Right now I'm finishing a cycle of tests across different business types. One archetype takes 30–60 cycles: test generation → error handling → optimization. I hope to have a beta version ready this spring. Though it's already spring 🤦♂️
Have a great one — for those of you whose Telegram is working!
#Upgrowplan