AX Consulting for Engineering Companies and Research Labs

AMORPH organizes the technical documents and lab records that engineering companies and research labs have accumulated into an ontology, and develops in-house AI, tailored to each client, that shows the source text and its location with every answer. An ontology is a knowledge structure that defines in advance the concepts used in the work and the relationships between them.

AMORPH is an AI software company based in Seoul. This page covers, in order, the services our AX (AI transformation) consulting provides, the companies we help, the principles behind the AI we build, what we do and do not take on, and what we have built so far.

What does AX consulting involve?

AX consulting means reviewing a company's work and materials, deciding where and how to use AI, then building the software that plan calls for and connecting it to the work. AMORPH's AX consulting starts with advisory work that organizes operational knowledge into documents and procedures, and extends to developing, supplying and maintaining AI software that answers from that knowledge.

AX (AI transformation) means changing work procedures that have already been digitized so that AI can judge and handle them.

Our services

AMORPH provides four services: knowledge organization advisory, technical document AI development, AI agent development and maintenance.

Knowledge organization advisory

We organize scattered operational knowledge into documents and procedures, and decide with the client which facts become company records and who verifies those records.

Technical document AI development

We develop in-house AI tailored to the client. It reads technical documents, proposes what to record with the quote and its location in the source attached, and records only what the person in charge has verified.

AI agent development

An AI agent is a program that combines a language model (an AI model that reads text and generates what comes next) with tools and rules, and carries out multi-step tasks on a person's request or at a set time or event. AMORPH develops AI agents fitted to the client's workflow, and designs them so that numerical comparisons and calculations are handled by predefined procedures rather than by the AI. Designing everything in an agent other than the language model in this way is called harness engineering.

Maintenance

We maintain the software we supply.

Which companies does AMORPH help with AX?

We help engineering companies and research labs that deal with many technical documents where figures and conditions are intertwined, such as specifications, test reports, technical reports and lab records. A specification states the conditions a product or material must meet, and a test report records the results of testing a product or material.

We are also preparing an industry-academia program to carry out joint projects with universities and government agencies. These are research projects that companies and universities carry out together. Work for entertainment companies, such as vocal conversion, is covered separately under Sound AI.

The principles behind the AI we build

The AI that AMORPH builds shows the source text and its location with every answer, records only what the person in charge has verified, and never makes up values without evidence. We first applied and tested these three principles in a knowledge AI prototype built by the AMORPH founding team. The prototype is software made to answer from documents and to record only what a person has verified.

First, every answer shows the source text and its location. For documents made with the Hangul word processor, the prototype attaches a source location to each piece of evidence, down to the table cell, such as "Section 1, Table 2, Row 3, Column 2". It accepts a quote as evidence only when the quote appears word for word at that location in the source.

Second, only what the person in charge has verified is recorded. A candidate record is something the AI has read from a document and proposed to record, but that no person has confirmed yet. In the prototype, even a candidate record that passes the checks becomes a company record only after the person in charge presses the confirm button.

Third, values without evidence are never made up. When asked about a condition for which the documents report no experiment, the prototype does not guess from the value of a similar condition; it answers that no experiment has been reported. A figure from a paper's table passes as a candidate record only when it sits in the same row, the same column and the same experimental condition. The problems that arise when reading paper tables are covered in "Can AI Correctly Read the Figures in a Paper's PDF Tables?"

Are internal documents sent to external AI?

Only when you allow external models and actually select one. An external model is an AI model that another company provides over the internet. The prototype calls an external model only when external AI calls are allowed in the environment where the program runs and an external model is selected on screen. In that case, the conversation, the contents of uploaded documents and any recorded knowledge related to the question are sent to the company providing that model.

What AMORPH does and does not take on

AMORPH develops, supplies and maintains software. We do not take on equipment or process design, on-site wiring or equipment installation, commissioning, or performance guarantees. Commissioning is the trial run of installed equipment before it goes into actual operation.

What have you built so far?

From September to October 2026, the AMORPH founding team built a knowledge AI prototype of the process in which the AI reads documents and proposes candidate records, and the person in charge confirms them. Using public papers and test documents, we tested whether source location display, confirmation by the person in charge and the blocking of values without evidence work properly.

What happens if the AI records something wrong?

In the prototype, nothing is recorded until the person in charge has verified it. If a wrong value is found after recording, the original record is not erased; a correction is appended, and any answers recorded on the basis of that value are recalculated.

Can it handle Hangul (HWPX) documents?

Yes. The prototype reads paragraphs and tables in HWPX documents and attaches source locations to tables at the cell level. Users with files in the older HWP format are asked to re-save them as HWPX and upload them. HWPX is a file format that stores the paragraphs and tables of documents made with Hangul, a word processor widely used in Korea, in XML, a structured text format.

Will AX consulting automate all of our work?

No. AMORPH's goal is to automate entire workflows with AI, but the software we build today keeps a step where the AI proposes and a person verifies and confirms.