# Optimisation: Get more out of the AI you’re already using.

Official page: https://aitje.com/en/services/optimalisatie

> AITJE to improve your AI workflow, chatbot or agent. Optimise skills, rules, reusable code, tool calls, knowledge bases, RAG and models in a targeted manner.

Whether it’s a workflow, chatbot, internal knowledge search tool, voice agent or coding agent: if you’re already using AI, there’s often more to be gained. AITJE investigates where time, context and computational power are being wasted, and where results can be improved.

This may be found in an agent’s instructions and skills, the structure of your knowledge base, the tool calls or the model itself. AITJE draws up an improvement plan and can implement the chosen changes. This applies even if you or another party has built the solution, either in-house or via external services.

## Who it is for

- A workflow involving many calls or repetitive tasks
- A chatbot that lacks the right knowledge
- A voice or coding agent that needs to work better
- High costs or inconsistent results

## The approach

### Measuring and analysing

AITJE It tracks your current tasks: what context is provided to the model, what does the knowledge base find, and where do errors, extra calls or delays occur?

### Improvement plan

Targeted adjustments to instructions, code, knowledge, retrieval or models. Including priorities, expected impact and the required hours and costs.

### Select & customise

You choose which improvements are to be implemented. Once the scope and hours have been agreed, AITJE will implement the agreed changes.

### Retest

The same sample tasks are carried out again. Quality, speed, calls and costs are compared with the initial situation.

## Deliverables

- An overview of your current operations and consumption
- An improvement plan for context, knowledge, code and models
- Priorities with expected impact, hours and costs
- After implementation: a comparison based on the agreed test tasks

## Prices and conditions

Fixed hourly rate. All prices exclude VAT.

For research and the improvement plan. Changes to the workflow, knowledge base or models are quoted separately, either by the hour or as a project. You give your approval in advance for the work to be carried out.

### Guardrailcheck

Approximately four hours for a single defined workflow: checking boundaries, human oversight, error handling and logging, and conducting targeted testing. You will receive the findings; any adjustments will be quoted separately.

From €299. All prices exclude VAT.

## Agreed separately

- Implementation of improvements (separate, at your discretion)
- External licences, API costs, hardware and server capacity
- A guarantee of a precise technical or financial result

## Frequently asked questions

### Can AITJE improve my existing AI?

Yes. ‘AITJE’ examines how your current workflow, chatbot, knowledge search tool, voice agent or coding agent works. The improvement plan may cover instructions, skills, code, tool calling, the knowledge base, RAG or model selection. You choose which adjustments ‘AITJE’ carries out.

### Even if another party has built the solution?

Yes, provided AITJE is given sufficient information and access to carry out the work responsibly.

### How do skills, rules and documentation help an agent?

They make the working method explicit: which steps an agent must follow, which code and tools already exist, and how the codebase is structured. ‘AITJE’ can organise instructions, Markdown files, skills and the folder structure more effectively. Reusable code and targeted tool calls prevent the model from having to figure everything out or generate it from scratch for every task.

### Can AITJE reorganise my knowledge base and vector database?

Yes. ‘AITJE’ can restructure documents and categories and improve the organisation of text sections, dates, priorities and access rights. The embedding model, index and search method can also be customised. If a different embedding model is used, the relevant documents are reprocessed and re-indexed. It is then tested whether RAG retrieves knowledge that is more relevant to your queries.

### What if the knowledge base and the questions use different languages?

AITJE It checks whether the embedding model adequately supports the languages used and whether the search query could be better formulated or translated. A multilingual embedding model or a different search approach may help. The choice is tested using your queries and documents, ensuring that the meaning and the correct source are retained.

### Does ‘AITJE’ also use the terms ‘fine-tuning’ and ‘ablation’?

Where appropriate for the task and the model. Fine-tuning uses examples to further train the task behaviour. Ablation alters a model’s rejection behaviour using available weights; this may be relevant when legitimate requests are unnecessarily rejected. It does not add any up-to-date business knowledge. ‘AITJE’ compares a modified version with the original version against the agreed tasks and checks.

### How do I know if an optimisation really makes a difference?

The same types of tasks are assessed both before and after the adjustment. AITJE looks at the usability of the result, speed, number of calls, resource usage and correction work. Fewer tokens or a cheaper model is only useful if the solution continues to meet the agreed quality standards.

### Is a Guardrailcheck a certification?

No. It is a practical technical assessment, not a legal judgement or a guarantee that a system will never make mistakes.

## Contact

[Contact AITJE](https://aitje.com/en/contact) — contact@aitje.com