We Started With a Simple Question
Can machines really understand markets the way experienced traders do? After years of watching algorithms miss what seemed obvious to the human eye, we decided to find out. That curiosity turned into something bigger than we expected.
How twinklecore Actually Happened
Back in 2019, three of us were sitting in a cramped Bangkok office, frustrated with how most financial models worked. They were either too rigid or tried to predict everything and failed at most of it. We kept asking ourselves: what if we could teach systems to recognize patterns the way experienced analysts do?
The breakthrough came from an unexpected place. Instead of trying to predict exact market movements, we focused on teaching neural networks to identify relationships between seemingly unrelated financial indicators. Kind of like how you might notice that your local coffee shop gets busier before stock market rallies because traders need more caffeine.
By 2022, we had developed training methods that actually worked in real conditions. Not perfect, never perfect. But reliable enough that people started asking us to teach them how we did it. So we shifted from building tools to building knowledge.
What Actually Matters to Us
These aren't corporate buzzwords we picked from a marketing handbook. They're the principles we argue about during long project meetings and the standards we hold ourselves to when nobody's watching.
Honest Limitations
We tell students what deep learning can't do before we tell them what it can. Markets are complex, unpredictable, and sometimes completely irrational. Anyone promising you guaranteed results is selling snake oil.
Real-World Testing
Every technique we teach has been tested with actual financial data, not just academic datasets. We've seen what works when servers crash at 3 AM and what fails when market conditions shift unexpectedly.
Continuous Adaptation
Financial markets evolve constantly, and so do our methods. What worked brilliantly in 2023 might be obsolete by 2025. We update our curriculum based on current market realities, not outdated textbooks.
Dr. Kasem Thongchai
Lead Instructor & Research Director
Kasem spent twelve years building trading algorithms for institutional investors before joining us in 2021. He got tired of explaining why his models failed during market crashes, so he started researching more robust approaches.
His teaching style is direct, sometimes brutally honest, but students appreciate that he shares his failures as openly as his successes. He believes the best way to learn deep learning is by understanding where and why models break down.
Our Teaching Philosophy in Practice
We don't believe in memorizing formulas or following step-by-step tutorials. Financial markets punish predictability, so we train people to think critically about model design and recognize when standard approaches won't work.
Start With Messy Data
Textbook examples use clean, pre-processed datasets that look nothing like real financial data. We throw students into the chaos immediately: missing values, conflicting timestamps, data from multiple exchanges that don't quite align.
It's frustrating at first. But that's the point. You learn more from spending three hours debugging a data pipeline than from running perfect code that someone else wrote. By the time students build their first working model, they understand why data quality matters more than algorithm complexity.
Build Models That Fail Gracefully
The difference between academic deep learning and financial deep learning? In finance, your model needs to know when it doesn't know something. We spend significant time teaching uncertainty quantification and confidence intervals.
Students learn to build systems that say "I'm not confident about this prediction" rather than producing precise numbers that turn out to be completely wrong. This approach doesn't make for impressive demos, but it's what actually works when real money is involved.
Our Next Program Starts September 2025
We're accepting applications for our intensive six-month program in deep learning for financial analysis. It's challenging, demanding, and requires serious commitment. But if you're ready to move beyond surface-level understanding and build real expertise, we'd like to hear from you.