LLM Inference Engine
View code →High-performance inference pipeline optimized for deep AI models with custom quantization and batching strategies.
PythonCUDAvLLMPyTorch
In progressBuilding minimal systems at the intersection of physics, mathematics and intelligent machines.
Amine B.H.T
About
Engineering student at l'École polytechnique de Bruxelles, specializing in applied physics and applied mathematics. Drawn to problems where fundamental science meets practical systems.
Currently focused on artificial intelligence and large language models, quantitative finance and trading systems and autonomous aerial robotics including FPV and unconventional drone architectures.
I build things that work with minimal complexity.
Areas of Work
AI / ML
- Fine-tuning & inference of small LLMs (3–7B)
- Retrieval-augmented generation systems
- Computer vision pipelines (YOLO)
- Model training, evaluation and deployment
Trading
- Market structure & prediction systems
- Algorithmic strategies and arbitrage
- Backtesting and simulation (Python)
Engineering
- FPV & autonomous drone systems (10+ years)
- Flight stacks & RC systems (Betaflight, EdgeTX)
- Physics-based modeling (nuclear, thermal)
- Thermal sensing & signal interpretation
- Systems programming (Python, Rust)
Building
Autonomous Drone Controller
View code →A signature project: custom flight controller with real-time sensor fusion, path planning and obstacle avoidance. Built on 10+ years of FPV experience.
C++EmbeddedBetaflightROS
Signature projectQuantitative Trading System
View code →Algorithmic trading framework with real-time market data processing, backtesting engine and risk management.
PythonNumPyPandasWebSocket
PlannedPhysics Simulation Engine
View code →Physics simulator designed specifically for ML training environments, prioritizing speed and scalability over general-purpose accuracy.
RustWGPUWASMWebGL
PlannedVision
Some ideas I find worth exploring:
The convergence of physics-based simulation and learned models will redefine what we consider 'intelligence.' Systems that understand physical constraints from first principles, rather than pattern-matching from data, will outperform on edge cases that matter.
Markets are information systems. The edge isn't in more data—it's in faster, more accurate models of how information propagates and transforms into price.
Drones are underrated as platforms for autonomous systems research. Constrained compute, real-time requirements and unforgiving physics create ideal conditions for developing robust AI.
The future belongs to those who can build complete systems with minimal resources. Complexity is a liability. Simplicity that performs is the goal.
Education
Ongoing
Engineering Degree
l'École polytechnique de Bruxelles — Applied Physics & Applied Mathematics