90-minute hybird session at Innovation Hub Tirana
The session, “A Gentle Intro to Deep Learning,” was the result of a powerful collaboration between Albanians in IT and AI Albania, two communities united by a shared mission: democratizing access to cutting-edge technology education in our region.
The session was held at Innovation Hub in Tirana’s Pyramid.

Expertise + Community
Besart Shyti, joining us remotely from London. As a former Machine Learning Engineer at Meta (Facebook), he worked on what he described as “the biggest deep learning system” at the company, the recommendation engine that determines what billions of users see in their feeds. Now, as co-founder of the Machine Learning Institute in London, he’s dedicated to making these complex concepts accessible to the next generation of AI practitioners.
But credentials alone don’t make a great teacher. What made this session extraordinary was Besart’s ability to strip away the mystique surrounding deep learning and reveal the elegant simplicity underneath.

Interactivity *
Besart made the session feel very much like a conversation.
Throughout the talk, he paused to ask the audience questions, for example:
““How can we represent this text as numbers?” “
— a question at the heart of AI systems.
He gave the group five minutes to discuss it in small groups. Ideas started flowing: some talked about keywords, others thought about word length or patterns in the text.
When the timer went off, Besart listened, inviting people to speak up before breaking down how embeddings and Word2Vec work.

“Deep Learning = Addition & Division”
“Deep learning might sound very complex… but all we did so far is plus and division. You will not even see a minus. You will see multiplication, but really multiplication is addition multiple times.”
This is a fundamental reframing. As Besart explained, “We are able to do something magical out of just very simple operations.” The “magic” of ChatGPT, Google Search, and recommendation systems comes not from complex mathematics, but from performing these simple operations at an incredible scale.
What loop happens in Google search
Besart revealed how search engines continuously improve through user behavior. Using Microsoft’s Bing dataset as an example, he explained:
“What Microsoft has done is they taken a query from a real user and they return you all the 10 documents that they gave to the users… the user clicked on this document here… You give this data set to Google every single day.”
Every time you search and click on result #3 instead of #1, you’re teaching Google that their ranking was imperfect. This continuous feedback loop is what makes modern search so powerful, and it’s happening invisibly, billions of times per day.

ChatGPT’s 2013 DNA
“What we did before was literally ChatGPT. If instead of the question being introduced in the middle, I put my question at the end… we have now something called a transformer decoder and transformer decoder is the other name for GPT.”
The core algorithm powering ChatGPT—predicting the next word in a sequence—is fundamentally the same as Word2Vec’s approach of predicting missing words. The breakthrough wasn’t algorithmic innovation alone, but computational power: “The algorithm was invented more than a decade ago. The only difference is that we’re waiting for GPUs to become incredibly powerful.”

RAG is cool
Besart connected it directly to today’s market realities. When discussing RAG (Retrieval Augmented Generation) systems—the AI document search solutions that companies are paying consultants significant sums to implement—he shared a telling anecdote:
“Someone came for an interview… I was asking them, what is a dot product?… He said, ‘I prefer building RAG systems’… Then I asked, “Well, how does a RAG system work? How do you know if a document is good compared to a query?”
The punchline revealed the gap between market hype and technical understanding: “What he was doing was exactly dot product, but just didn’t know the name.” This resonates strongly while navigating an industry where AI buzzwords often mask fundamental concepts.
Why this matters
The session wasn’t just about individual learning, it represented something larger. AI Albania, led by researchers and practitioners committed to building the country’s first comprehensive AI excellence center, has identified a critical need: bridging the gap between global AI developments and local expertise.
As the organization works across three key areas—AI literacy for all, research and education, and governance and policy—events like this demonstrate the power of combining local community building with international expertise.
Albanians in IT, with its focus on connecting diaspora expertise with each other and local talent, provided the perfect platform for this knowledge transfer. The hybrid format—Besart presenting from London while participants gathered at Innovation Hub Tirana—exemplified how geography is not necessarily a limit for access to world-class education from Albanians to Albanians.

Key takeaways for the community
What made the session particularly powerful was Besart simplifying fundamentals.
The five major insights participants gained were:
- Simplicity underneath complexity: Modern AI relies on basic mathematical operations scaled massively
- Continuous learning systems: Every user interaction trains AI systems in real-time
- Historical continuity: Today’s breakthroughs build on decades-old foundations
- Relative vs. absolute learning: AI learns relationships and comparisons, not absolute truths
- Market-ready applications: Understanding fundamentals enables building valuable systems
Additionally, as Besart noted during the wrap-up, this was his first time presenting to a predominantly Albanian-speaking audience, and the mutual enthusiasm suggests it won’t be the last.
What’s next?
The session has validated the model of combining local community organization with diaspora expertise. As both Albanians in IT and AI Albania continue building their programs, they now have proof that complex technical education can work in this format.
For Albania’s broader tech ecosystem, the session represents a template: Albanian world-class education is possible when communities organize effectively, leverage their networks thoughtfully, and focus on genuine value creation over hype.
The recording and materials from the session are available to community members. To join future sessions and stay connected with both communities, follow Albanians in IT and AI Albania on their respective platforms.
About the communities:
Albanians in IT connects Albanian tech professionals globally, fostering knowledge sharing and career development through events, mentorship, and community building.
AI Albania is building Albania’s first comprehensive AI research and education center, working across AI literacy, research, and policy development to prepare the country for the AI-driven future.
Innovation Hub Tirana, located in the Pyramid of Tirana, provides space and support for tech events, startups, and community gatherings that drive Albania’s digital transformation.

Leave a comment