Building systems at the intersection of AI, mathematics, physics and software engineering.
I’m a student and independent developer interested in understanding things from first principles and turning that knowledge into working systems.
My main project is EventHorizon-AI — an independent research and engineering project focused on building forecasting systems and validating them with rigorous statistical methods.
|
EventHorizon-AI is my long-term research and engineering project. The goal is simple: Build models. Test them honestly. Find out what actually works. The project explores forecasting across different domains, from financial time series to retail demand.
|
|
A model being accurate is not enough.
A statistically significant result is not automatically economically valuable.
And a technically validated system is not automatically a product people need.
That distinction is now part of the way I approach every project:
Model → Statistical Evidence → Real-World Constraints → Product Value
A BTC/USDT short-horizon directional prediction system.
The research pipeline detected a statistically significant directional edge at the tested 5-second horizon.
However, after accounting for exchange fees, the strategy was economically unviable.
That result became an important research case study:
Finding an edge is not the same as finding a profitable system.
A retail demand forecasting system using machine learning and temporal features.
The system was evaluated against a seasonal baseline using a rigorous time-series validation pipeline.
WAPE
- Baseline: 44.4%
- Model: 30.9%
- Improvement: ~30%
The technical model is validated; the current challenge is determining where forecasting creates enough operational value to become a useful product.
An open-source Python toolkit for rigorous time-series validation.
It contains reusable methods for:
- Walk-forward splits
- Block bootstrap
- Gap bootstrap
- Permutation testing
- WAPE
- MASE
The goal is to make it harder to accidentally mistake temporal structure, autocorrelation or leakage for genuine predictive evidence.
- Calculus
- Linear algebra
- Complex numbers
- Probability and statistics
- Mathematical modeling
- Classical mechanics
- Special relativity
- General relativity
- Quantum mechanics
- Quantum computing
- Machine learning
- Time-series modeling
- Scientific computing
- Algorithms
- Systems programming
- C++
- Python
Languages
Python · C++ · JavaScript · HTML · CSS
Machine Learning & Data
LightGBM · PyTorch · NumPy · Pandas · Scikit-learn
Scientific Computing
SciPy · Matplotlib
Development
Git · GitHub · VS Code · GCC
I care about the difference between:
"It works on my machine."
and
"There is evidence that this actually works."
So I try to:
- Form a hypothesis.
- Build the smallest system capable of testing it.
- Validate it against appropriate baselines.
- Look for leakage and statistical artifacts.
- Quantify uncertainty.
- Document failures.
- Test the system against real-world constraints.
- Only then decide what is worth building further.
I'm currently balancing three long-term goals:
Mathematics & Physics
Understanding the mathematical foundations behind modern physics and computation.
Engineering
Building increasingly sophisticated software and machine-learning systems.
Entrepreneurship
Turning technically rigorous ideas into products that solve real problems.
My long-term interests lie particularly at the intersection of AI, physics and computation.
Rigor over hype.
Evidence over assumptions.
Understanding over memorization.
Research over guesswork.
Proof, not promises.
Building. Testing. Learning. Rebuilding.