Back to portfolio
Case Study

MAIstr

An AI knowledge manager your know-how cannot walk out of

View live
MAIstr - the interface for training an AI agent by uploading video, audio, a document or text

About the project

MAIstr grew out of a real problem in our family business. The knowledge of experienced people lives in their heads, in notes and in e-mail threads, and when someone leaves, the way things are done leaves with them. MAIstr lets know-how be captured in four ways: upload a document, record it by voice, film a video or simply write it down. The AI processes the input, extracts what matters and stores not only the information itself but also its connections to what is already in the system. That is what makes answers accurate and relevant rather than merely similar to the question. An employee then asks in plain language and gets an answer immediately, at any time. Data security is treated as a priority, because what goes into the system is the most valuable thing a company has. The project is being built under my direction and is in a pilot phase, onboarding its first customers.

Project information

Client

Own product

Platform

Web application

Status

In development

Development time

Ongoing, pilot phase

Year

2026

Technologies used

Check out the stack I used for this project

Frontend

ReactVite

Backend

PythonGeminiClaude

Database

QdrantPostgreSQL

Key features

Knowledge captured four ways: document, voice, video, text

AI processing of inputs and automatic knowledge base creation

Connections between pieces of information kept, not isolated snippets

Questions in plain language, answers immediately and at any time

Specialised agents for individual areas of the company

An overview of unanswered questions as input for filling gaps

Vector search across the knowledge base

A focus on data security and control over where data goes

Challenges

  • 1

    Knowledge is scattered across a company and often exists only in people's heads

  • 2

    Capturing know-how without costing an expert hours of writing

  • 3

    Similarity search returns fragments stripped of their context

  • 4

    The most valuable company data goes in, so security cannot be an afterthought

  • 5

    Combining several models and data stores into one working whole

Solutions

  • 1

    Four capture paths including voice and video: describing a procedure out loud is enough

  • 2

    The AI processes, sorts and prepares the input into the knowledge base itself

  • 3

    Storing connections between pieces of information, so answers keep their context

  • 4

    Data security designed in as a product priority from the start

  • 5

    Qdrant for vector search and PostgreSQL for structured data

Results

Know-how stays with the company even after an experienced employee leaves

Onboarding relies on answers instead of digging through documents

Answers keep connections across sources rather than text similarity alone

The product is running in a pilot with its first customers

Development timeline

From analysis to deployment - how the project evolved

ongoing

Validating the problem

The problem named in the family business, interviews confirming other companies face it too

iterative

Knowledge base prototype

AI processing of inputs, designing how connections are stored, vector search in Qdrant

iterative

Application and agents

React and Vite interface, Python backend, specialised agents and the training workflow

current

Pilot operation

Onboarding first customers, feedback from real use and adjusting priorities

Have a similar project?

I'll be happy to help you build your application. Feel free to contact me for a free consultation.

Contact me