Applied AI
AI takes over the tasks that eat up your day.
I am a bioprocess engineer and I apply AI to the daily work of technical companies. I start by understanding how you operate, build what really saves time, leave it running in your environment and train your team to use and maintain it.
In short
Dani Canton implements applied AI in biotech and industrial companies and trains their teams to use it. The work is organised by task and by result, not by tool: automating market monitoring and competitive studies, assistants over internal documentation, support for technical and validation documentation, and analysis of process and laboratory data. He is a bioprocess engineer with more than 8 years in GMP environments, so he understands the process the AI is applied to. He is based in Barcelona and works with clients in Spain and abroad.
The problem
I do not come from software. I come from process.
Most AI projects do not get complicated because of the technology. They get complicated because they are built without understanding the process they are applied to, or because they stop at a brilliant pilot that nobody connects to the real work. Six months later there is a tool used by one and a half people.
I have spent years translating business needs into processes that work in the plant. I apply the same judgement here: first I understand how you operate and what your data looks like, then I build what really improves it. And I tell you what makes no sense to automate, which is usually the most useful part of the conversation.
What I implement for clients is what I use myself every day. I do not sell a technology I have only read about.
What I do
By task and result, not by tool.
Market monitoring and competitive studies
Automating the tracking of sources, trials, regulatory updates and news that is done by hand today, with every data point linked to its source. A monthly study that takes a couple of days becomes a short review.
An assistant over your internal documentation
Querying protocols, reports and documents in plain language, in your environment, with confidential data kept in-house. No more searching by hand for what is already written.
Technical and validation documentation
Support with drafts of reports, summaries and process and validation documentation (FAT, SAT, CSV). The AI does the heavy lifting of drafting; a person reviews and signs.
Process and laboratory data analysis
Cutting the time spent on repetitive analysis and report preparation that takes a whole day every time today.
Team training
Hands-on sessions on your people's real work, not a talk about concepts. You leave with one of your own cases working, not with notes.
Infrastructure and roll-out
Leaving the environment set up and accessible to the team, with shared tools and clear rules of use, so it stays alive when I am not around.
How it works
Five steps, and you decide at each one.
- Diagnosis
I learn how you operate and look at where AI adds value and where it does not. There are always more candidates than are worth doing.
- Prioritised use case
We pick one with a measurable result in hours saved or in work that no longer waits for you. This is where you decide whether we continue.
- Pilot
On your data and your real workflow, not on a pretty demo. This is the point where you see whether the idea survives contact with reality.
- Implementation
I leave it running in your environment, with its access rights, its limits and its documentation.
- Training
Your team learns to use it and maintain it. If you depend on me to keep it standing when we finish, the job was done badly.
Your data, your control
The three usual concerns, settled from the start.
- Traceable. Every conclusion links to its source. If the system does not have a piece of data, it says so and passes it to human review instead of making it up.
- Private. Deployments in your environment when the data cannot leave. NDA and governance from day one, not as a patch at the end.
- Under your judgement. AI proposes, people decide and sign. In a regulated environment there is no other way to do it.
The proof
Why me and not an AI consultancy.
- I know the sector from the inside. More than 8 years of bioprocess work in biotech, pharma and cosmetics, in environments where the technical and regulatory context is not a detail.
- I use this every day. My own work (process, documentation, client follow-up) runs on the same thing I implement. If it does not work for me, I will not build it for you.
- I have already done it outside my own house. Hands-on AI training for a diagnostics spin-off: two sessions with their team and, at the end, one of their real cases working (a monthly market study that used to take them two days).
- I start with what can be measured. One case with a verifiable result before proposing anything big.
FAQ
What people usually ask me
Does my data leave the company?
It depends on what is built, and it is decided with you before we start. When the data is sensitive or there is a confidentiality commitment involved, the deployment is done in your environment and the information does not leave. When external services are used, what is sent to them is limited and an NDA is signed. Either way you will know it in writing, not as small print.
Does this work in a GMP or regulated environment?
It works for the knowledge work around the regulated process: searching, summarising, drafting, preparing documentation and analysing data. The rule is that the AI does not sign or decide: it proposes, and a qualified person reviews and takes responsibility. Anything that becomes part of a system with GxP impact goes through validation, like any other software.
How long until there is a result?
The first use case takes weeks, not months, precisely because it is chosen to be small and measurable. If someone proposes a six-month project before testing anything on your data, be wary.
What if my team is not technical?
That is the normal case, and it is not a problem. The training is about each person's real work, with their documents and their tasks, not about how a model works inside. What is needed is to know your own work well, and they already do.
What happens when you finish?
It keeps working without me. It is handed over documented, with the access rights in your hands and someone on your team able to maintain and extend it. If you then want to keep building together, great, but out of interest and not out of dependence.
Do you automate any task?
No, and saying so early saves money. The tasks that automate well are repetitive ones, with a criterion that can be explained and an accessible source of information. The ones that depend on professional judgement, on a relationship with a person or on information that only lives in someone's head do not pay off, or turn out badly.
Contact
Is there a task you think could be automated?
Tell me about it and I will tell you frankly whether AI can handle it, what it would take and what is not worth touching.
Tell me your case