Marcus Biel.

AI software development and integration

AI development for the work your teams do every day

An AI feature can pull line items from a PDF order, find an answer in a service manual, or draft a reply to a customer. I build those features into the software your team already uses.

I start by asking about a specific task. What information does someone need, what should the software produce, and who checks it before it goes any further?

I’m Marcus Biel. I work with your team on those decisions and help build the feature.

What do you need AI to do?

Start with what goes in and what you need back. A purchase order calls for checked line items. A question about a product calls for an answer your team can trace to a source.

Extract information from documents

Your team regularly transfers information from PDFs, emails or forms to another application.

The application extracts the fields your team needs in a format your system can use. That might mean turning an emailed order into line items or pulling customer details from a form.

Staff can compare the values with the original, correct mistakes, and approve the transfer.

Discuss document processing

Find answers in your company documents

A single support question can send someone looking through a manual, a wiki, and a set of internal notes.

I build assistants that search the sources you choose and show the passages behind each answer. You decide which documents to include and who may read them.

With retrieval-augmented generation (RAG), the application finds relevant passages and gives them to the model to use when answering the question.

Explore AI knowledge assistants

Discuss a knowledge assistant

Example

Turn a PDF into an order your team can process

Sample order BN-1042 lists 12 mounting plates and six bearings. The first line reads correctly. On the second, the model reads the last digit of RL-18 as the letter B.

Compare the draft with the PDF, correct the item code, and see which values are ready to send to the order system.

Talk about your document workflow

From incoming PDF to approved order

  1. Receive the order The order arrives through a file upload, a shared mailbox, or another channel you use.
  2. Extract the details The AI reads fields such as the order number, line items, quantities and requested delivery dates.
  3. Check the results Staff see the extracted values alongside the document. The application checks required fields and agreed rules.
  4. Correct and approve A reviewer fills in missing details, corrects mistakes, and approves the order.
  5. Send the approved order The connected system receives the approved information.

Check the item code against the sample order

Compare the item code in the draft with the original order. One character needs a closer look. The demo uses fictional data and runs in your browser.

Original document

Open PDF
WERKSTATT NORD

Order

No. BN-1042 · Sample document

Line items on the original order
ItemCodeQuantity
Mounting plateMP-2412
Round bearingRL-186

Please deliver both items together.

Draft order

BN-1042
Mounting plateMP-24 · 12 units Find line 1 on the original
Confirmed
Round bearingRL-1B · 6 units Find line 2 on the original
Awaiting review

The last character could be a letter or a digit. Which code appears on the original?

Three ways to start your AI project

I can help you define the task, test it on real examples, or build it into your software. For each piece of work, I make clear what I’ll deliver and what’s in scope.

Define a use case

You’ve spotted a task that might suit AI. I help you decide whether it’s worth pursuing and where to start.

I look at the existing process, the available data and the desired result.

You get a short plan covering the inputs, the output your team needs, the systems involved, and what to build first.

Run a focused AI pilot

You want to know how well AI can handle a particular task with your own data.

I build a working version and test it with agreed examples.

What’s included in a pilot
  • the specific task and the example material used
  • the expected output
  • the agreed data sources and connections
  • criteria for checking results
  • how to handle ambiguous or incomplete output
  • an evaluation and recommendation for the next steps

You can see where the feature works, where people still need to step in, and whether another round of development is worth it.

Example of a pilot evaluation

Integrate an AI feature

You already know what the feature should do and how it will fit into the application.

I build the agreed feature into your application, including the user interface, system integrations, permissions and processing of results.

I plan rollout and support with the people responsible for the system.

See system integration in a project
Discuss your AI project

Integrate AI into Java applications you already run

Your team can use AI without leaving the Java application where they already handle orders, tickets, or documents.

I look at the codebase with your developers and choose where to add the feature.

I’ve worked with Java since 2001. I’ll design the integration with you and write the code alongside your team.

Decisions to make before development

  • where people access the feature
  • what data the feature needs
  • how the result is displayed and processed
  • which permissions it must enforce
  • what users see when a request takes longer than expected or fails
  • how to test and deploy the extension

How I check the answers

See a sample pilot evaluation

Sample order BN-1042. Line one matches the PDF: 12 units of MP-24. Line two has the right quantity, six, but the model suggests RL-1B. The source reads RL-18.

What happens next. A person must approve the second line before the order reaches the business system. A pilot with your documents would show the types of errors and how often your team needs to intervene.

Check the sample order

Before writing code, I ask what a useful result would look like to you. For a purchase order, that might mean the right quantities against the right item codes. For a knowledge assistant, it means an answer you can verify in the cited documents.

I test with examples from your team’s work, including the unusual cases people still need to handle.

I check whether someone can spot and correct a mistake, and what they see when the model has too little information to answer.

Those checks are part of the feature your team will use after launch.

Who builds your AI integration?

I’m Marcus Biel. I design the integration and help build it, working with your developers or bringing in specialists where needed.

I work through technical decisions with the people who will use, run, and maintain the feature. They know which systems it needs to connect to and what happens when something goes wrong.

I work remotely with clients across Germany.

More about working with me
Marcus Biel

Frequently asked questions about AI development

What data do you need to get started?

Show me a typical input and an example of the answer or action you want.

For document processing, share typical documents and the fields you need. For a knowledge assistant, sample sources and real user questions are useful.

I’ll check with you first about which data I can use for development and testing.

Do we have to develop our own AI model?

Usually an existing model will do. I can connect a large language model (LLM) to your application and the data it needs.

I’ll use examples from your workflow to choose the model and the checks around it.

How is our company data handled?

I map out what data leaves each system, where it goes, and who can access it.

Before choosing a provider or hosting environment, we agree on what the application can store and who can see it.

How much does AI integration cost?

Cost depends on the task, the data, the systems I need to connect, and how thoroughly the result needs to be checked.

Once the scope of the first release is clear, I can quote for them or work within a budget for the first phase.

I also consider model usage, infrastructure and support. A pilot can help estimate those costs at your expected volume.

What happens after the pilot?

I review the results with you so you can decide whether to refine the feature or roll it out.

You can move ahead with integration, try another round of testing, or change direction. We agree on the scope before starting more work.

Where could AI save your team time?

Briefly describe today’s process and the outcome you want to achieve.

If you already have an application, sample documents or an idea, we can start there.

We’ll look at the task together and choose a first step you can test.

Prefer to talk first? Call +49 171 1151013 or message me on LinkedIn.

Talk to Marcus about AI