I am a self-driven, autonomous developer building at the boundary of Deep Reinforcement Learning (RL), Quantitative Finance, and Asynchronous AI Orchestration Loops. My primary sandbox involves constructing high-frequency self-evolving trading engines, complex intermarket feature pipelines, and bulletproof security gateways.
- š¬ Driven entirely by deep, continuous self-directed experimentation.
- š”ļø Obsessed with strict quality control, automated regression suites, and production container hardening.
- š§¬ Believer that clean, modular math always wins against market chaos.
| Operational Layer | Tooling & Infrastructure |
|---|---|
| Deep Learning & RL | PyTorch ā¢ Stable-Baselines3 (SAC / PPO) ā¢ Gymnasium ā¢ Walk-Forward Bars |
| Data & Relational Graphs | Pandas ā¢ NumPy ā¢ NetworkX (Topological Path Reasoning) ā¢ ChromaDB ā¢ Supabase |
| Core Core Engines | FastAPI (ASGI) ā¢ AsyncIO ā¢ Flask (Factory Pattern) ā¢ Gunicorn |
| MLOps & Assurance | GitHub Actions ā¢ Ruff ā¢ Black ā¢ Pytest (Async/Mock Frameworks) |
| Deployment Security | Multi-Stage Docker Builds ā¢ Secure Non-Root Privilege Boundaries (USER) |
š¤ xau_rl_engine_v2
A completely decoupled deep-RL algorithmic trading network. Features a PyTorch Temporal Attention Network (Oracle) using learnable
[CLS]token configurations to forecast class directional probabilities, passing filtered streams down to a Stable-Baselines3 Soft Actor-Critic (SAC) framework for automated execution risk-management.
š xau-ppo-lstm-engine
Continuous-space Deep Reinforcement Learning framework leveraging Proximal Policy Optimization (PPO) paired with a recurrent LSTM backbone tracking layout to balance live asset allocations across dual-stream spot markets (XAUUSD / DXY).
A deep neural pre-trade firewall engine classification system. Ingests sequential rolling multi-candle OHLCV tensor windows to isolate impending volatility expansions, backed by strict custom backpropagation parameter gradient assertion suites.
An enterprise ASGI gateway featuring an autonomous asynchronous ReAct routing loop. Seamlessly fuses semantic vector search (ChromaDB) with topological graph network mapping (NetworkX) and a secondary LLM-as-a-Judge validation proxy layer.
High-throughput edge routing gateway microservice designed to ingest real-time Meta WhatsApp webhooks, executing instant 200 OK handshakes under strict latency constraints while safely offloading GraphRAG processing loops to background worker threads.
Every script hitting this profile goes through an unyielding pipeline baseline:
ā” RUN: ruff check . # Zero code smell syntax policy ā” RUN: black --check . # Deterministic style layout parity ā” RUN: pytest --asyncio # Isolated mocking & regression coverage ā” BUILD: docker build --no-cache # Multi-stage compilation & non-root lockdown