Assistants that answer using your documents, automation that gets repetitive work out of the way, models that run locally when data must not leave the company. I design and build systems based on artificial intelligence starting from a concrete problem, not from this month’s novelty.
When it makes sense to bring artificial intelligence into a company
Repetitive tasks are eating up hours
Quotes, reports, product sheets, translations, sorting requests. If a task always follows the same pattern, it can almost always be automated entirely or in large part, leaving people the final check.
The information exists, but nobody can find it
Manuals, procedures, catalogues, project history. An assistant connected to internal documents answers in natural language and points to where the answer came from, so it can be verified.
Data cannot leave the company
Professional firms, healthcare, research and development: in many cases sending documents to an external service is not an option. Language models can run on a computer or server you own, with no outbound connection.
AI is already in use, but everyone does it their own way
Someone uses a chatbot for emails, someone else for copy, nobody knows what really works. The step needed is from individual use to shared workflows, with clear rules and measurable results.
What I can do for you
Analysis and strategy
- Mapping processes and choosing where AI pays off (and where it does not)
- Comparing cloud services and local models: costs, limits, confidentiality
- Working prototype on a real case, before investing
- Working alongside your legal adviser or DPO on the technical side of GDPR and the AI Act
Assistants and agents
- Assistants that work on company documents, catalogues and archives
- Long-term memory based on knowledge graphs
- Agents that carry out tasks using tools: files, databases, APIs, management software
- Voice interfaces: listening, spoken replies, handling interruptions
Automation and workflows
- Orchestration of several specialised agents on the same task
- Content production and optimisation: copy, translations, images, SEO
- Integration with WordPress, WooCommerce and tools already in use
- Bespoke scripts and micro-software, mostly in Python
Quality and control
- Repeatable tests to compare models and spot regressions
- Checks written in code, not left to the model’s goodwill
- A log of what the system did and why
- Training for the people who will use it every day
Local AI: when the models run in-house
A language model does not have to live in someone else’s cloud. Today there are open models that run on a good graphics card or a small server, in the office. The advantage is obvious: documents do not leave the company, the cost is predictable and the system keeps working even without the Internet.
There is a flip side, and it is better to know it beforehand. A small model goes wrong differently from a large one: it may claim to have done something without doing it, or fill a gap with a plausible invention. That is why a local system has to be designed around the model’s limits: controlled memory, automatic checks, tasks split into small steps. It is the part of the job I know best, because I practise it every day on my own systems.
Often the right solution is mixed: local models for confidential data, cloud services for tasks that need more power and involve no sensitive information.
AI agents and orchestration: how I work every day
A good part of my work today goes through AI agents collaborating with one another. I design the architecture and decide what needs doing; different agents write the code, review it looking for security holes, measure performance, check SEO and copy. Every result is verified before being accepted.
I say it plainly because it is a strength, not a secret: I do not write every line by hand. I design systems and build them with AI assistants, and I can explain the reason behind every choice. It is the same method I bring into companies: artificial intelligence pays off when someone gives it direction, limits and oversight.
The interest did not start with the chatbot craze. My thesis, in 2013, was on the semantic web and ontologies: the same way of organising knowledge that today lets assistants remember and reason over a company’s data.
How I work
01
Listening
You tell me how you work today and where time is lost. Without knowing the process, any tool is a gamble.
02
Test on a real case
Together we pick a small, measurable task, and I build a prototype that works on your data.
03
System
If the test holds up, it becomes a stable tool: integrations, checks, activity log, documentation.
04
Handover
Training for those who use it and support over time. The goal is for the system to remain yours, not for you to depend on me.
Projects and research
Local AI assistant with graph memory
A voice assistant that runs entirely locally, remembers over time and is put to the test with a model evaluation kit. View the project
WP Translate Engine
WordPress plugin that translates an entire site with AI models, including local ones, and takes care of the SEO side. View the project
Frequently asked questions
How much does it cost to introduce artificial intelligence in a small company?
It depends on what it has to do. You almost always start from a limited case, so the initial outlay is contained and the result can be measured straight away. There are three cost items: design and development, hardware for local models or the fees for cloud services, and maintenance. After a first conversation I will tell you what is really needed and what is not.
Is a cloud model or a local model better?
The cloud offers the most powerful models and no hardware to manage, but data passes through an external provider and cost grows with use. Local keeps everything in-house at a fixed cost, with less powerful models that need more guidance. The choice depends on how confidential the data is and how hard the task is; often both are used.
Is my company’s data used to train the models?
It depends on the service and the contract: the professional plans of the main providers normally exclude using data for training, free ones often do not. With a local model the problem does not arise, because nothing leaves your network. Checking where the data ends up is part of the initial analysis.
Do I need in-house technical staff to use these tools?
No. A well-made system is used like any other program: a window to type in, a button, a folder it reads files from. What you do need is a reference person who knows the process and flags when something does not add up.
What changes with the AI Act?
The European regulation on artificial intelligence sets different obligations depending on the risk of the use made of it, as well as rules on transparency and staff training. I am not a lawyer: I handle the technical side, that is, knowing which systems are in use, where the data goes and what is logged. It is the basis of any compliance work, and I prepare it working together with your adviser.
You may also be interested in
WordPress consulting: performance and security
Bespoke websites and e-commerce
Partner for agencies and studios
Let’s talk about AI in your company
Tell me what you do and where you would like to save time: I will tell you frankly whether artificial intelligence is the right tool.
Email
valentino.chiarla@gmail.com
Where
Turin, Italy. I work with companies, agencies and studios in Piedmont, in Milan and across the rest of Italy, in person or remotely.