I build software to investigate complex systems — from intelligent investment systems and cognitive architectures to artificial observers and the mechanisms underlying intelligent behavior.

My work sits at the intersection of:

Software Engineering · Artificial Intelligence · Quantitative Research · Cognitive Science · Computational Neuroscience

I am particularly interested in one question:

How does an intelligent system build an internal model of a world it cannot fully observe?

Cognitive Observatory
An experimental laboratory for studying artificial observers, internal models of reality, memory, attention, prediction and uncertainty.

The core idea is simple:

REAL WORLD
    ↓
PARTIAL / NOISY OBSERVATION
    ↓
OBSERVER
    ↓
INTERNAL MODEL
    ↓
DECISION / ACTION
    ↓
WORLD

The experiments measure how an observer's internal model diverges from the actual state of the world as its cognitive architecture changes.

Observador Lab
Mechanistic interpretability experiments with small language models.

The central question:

Does a model's explanation of its own behavior correspond to the mechanisms actually producing that behavior?

Experiments focus on models such as GPT-2 and Pythia using local, reproducible analysis.

DriftLab
An experimental platform combining browser-based cognitive tasks with computational modeling.

Current work includes:

• Stroop
• Flanker
• N-back
• Reaction-time experiments
• Psychophysics
• Drift-Diffusion Models
• Decision threshold vs. processing speed analysis

The goal is to connect observable behavior with computational models of decision-making.

JV Investments

Private quantitative research focused on building intelligent systems for financial markets.

The broader objective is to combine:

• Market data
• Web-scale information
• Automated data collection
• Statistical analysis
• Machine learning
• AI-assisted decision systems
• Systematic trading

The philosophy is not to ask AI to predict the market, but to build systems capable of extracting, evaluating and combining information under uncertainty.

I also maintain smaller projects, experiments and prototypes across:

• Python
• C#
• Kotlin
• JavaScript / TypeScript
• React
• .NET
• Web technologies
• Data analysis
• Automation
• Machine learning
• Computational experiments

Not every repository is intended to become a product.

Some exist simply because the fastest way to understand an idea is to build it.

Artificial Intelligence
        │
        ├── Agent architectures
        ├── Mechanistic interpretability
        ├── Memory & attention
        ├── Internal representations
        └── Artificial observers
                │
                ↓
        Cognitive Science
                │
                ├── Decision making
                ├── Predictive processing
                ├── Psychophysics
                └── Computational neuroscience

Quantitative Research
        │
        ├── Financial markets
        ├── Automated data collection
        ├── Statistical systems
        └── Intelligent investment systems

17+ years building software systems across different technologies and domains.

My approach has gradually shifted from:

building software that solves a predefined problem

toward:

building experimental systems that allow me to investigate problems that are not yet fully understood.

That distinction matters.

I am interested in systems where the important question is not simply:

Does it work?

but:

Why does it work?

And eventually:

What does that tell us about the system itself?

• Artificial observers
• Cognitive architectures
• Mechanistic interpretability
• Memory and attention
• Predictive models
• Decision-making systems
• Quantitative finance
• AI-assisted research
• Computational neuroscience
• The limits of observation and self-modeling

🌐 natyari.com
💼 LinkedIn
🐦 X
📷 Instagram
💻 CodePen

Build the instrument. Run the experiment. Follow the evidence.