The problem
In a world where we live and work surrounded by devices that are constantly generating and releasing information, it is as difficult as it is important to know how to navigate this sea of documents of every kind that surrounds us. Doing so efficiently is key to being able to find what we're looking for. But when your day-to-day consists of working with databases, processing pipelines and visual reports for the airline industry, the challenge multiplies. We operate in a highly regulated environment, where information is complex, data volumes are massive and rigour is absolute.
In my personal case, I am also blessed with a remarkably privileged memory that lets me remember a random, useless detail from 20 years ago, yet is unable to retain the definition of a work process I've already needed four times in the last year.
Managing teams also adds another layer of complexity: when someone new comes to work with us, onboarding involves an enormous knowledge transfer. The documentation they need to be given about the client, the business and the technical stack is spread across:
• Official tool documentation (cloud services, dbt, Redshift, etc.).
• Official client documentation and regulations (airlines).
• Ad hoc processes and business rules that have been documented in different ways and by different people over time.
• Documentation we have gathered ourselves over time, with little tricks, system quirks and recurring SQL queries.
With all this, the second phrase I repeat most, after “let me look into it and get back to you”, is, without a doubt, “this has happened before, but I can't remember right now where or how it was solved”. And, until not so long ago, that meant navigating through the documents suspected of containing the information I was looking for.
The discovery: NotebookLM
I admit it was somewhat by chance, but one day I came across a tool that changed my life: NotebookLM. It was only to be expected that AI would, sooner or later, show up to help me in my aim of minimising the time spent on tedious tasks and, better still, freeing up the valuable space in my memory that until now had been wasted on knowing where certain things are.
This application, which is free, lets me upload my files of any type (PDF, Word, text, audio, image, video) or point it to URLs containing relevant information, so that I can then ask, through a chat with Gemini built into the platform itself, anything about those sources. What's more, when the AI responds, it indicates the source (and not just the file, but also the relevant paragraph) used to give that answer,so I can validate where the information comes from or pull on the thread, if I need to. This feature alone already solves a huge problem for me, but there's more.
The other side of the tool is based on generating content from the uploaded sources. And we're not talking about plain text: we can generate a presentation, a video or audio summary, an infographic or even a quiz. This opens up endless possibilities for consuming the information in our sources in a way that is more educational, more practical or, why not, more enjoyable.
⚠️ Note on the responsible use of AI: When using free, public AI tools like this one, it is vitally important not to share sensitive or confidential information. You should never upload real customer data, financial information or internal, confidential details of your clients' projects. We must always work with anonymised data, generic structures or public technical documentation.
Examples from day-to-day work
Querying a database
In the first example, I have the documentation for an airline's SQL bookings database, along with some reference SQL queries. I need to write a query against the database that lets me identify all passengers who did not show up for their flight over the last month. And I also need an explanation of the relevant fields used in that query.
Sure enough, NotebookLM has generated a fully working query for me, and then explained the relevant fields. Worth highlighting, as we mentioned, are the references to the original sources, something essential to ensure that the data we're extracting is correct according to the client's rules, as well as the strict practice of working only with anonimized and public data.
Onboarding and knowledge transfer
In the second example, let's say a junior profile joins the team and I need them to get familiar with our working methodologies quickly. I upload as sources the technical manuals, codestyle guides (for example, for dbt or Python) and the general documentation on the structure of our repositories.
As there is a lot of technical information and I need them to start with the key concepts of our architecture, I ask the tool for a mind map of the data flows.
To make sure they absorb the team's standards before touching real code, I ask the AI to generate quizzes on the different technical processes they will have to apply in their day-to-day work.
If the new team member prefers another format, I can even ask it to generate a video summary explaining the fundamentals of our pipeline, so they can watch it while they review the diagrams.
Conclusion: The value of a good tool used with judgment
As we have seen, the possibilities are as endless as the savings in time and costs. This approach adds enormous differential value to our work: what used to take hours of digging through wikis to understand a regulated data ecosystem, trace the origin of a field in a table or train a colleague on a complex architecture now becomes an agile and precise process. It lets us focus on building and optimising, instead of wasting memory on remembering where things are.
Although I've focused these examples on my own technical reality, the best part is that it is a fully cross-functional tool, one that can serve the data engineer who has to dive into a new system (and its database) just as well as the student preparing for an exam, the communications manager who has to produce extensive infographics, or the legal team that needs to audit extensive regulations.
