AI Systems Engineer · Full-Stack Developer · Open Source Builder

Building production-grade AI and ML systems that solve real-world problems

I'm an independent developer specializing in AI systems architecture, machine learning platforms, and RL optimization frameworks. My focus is building production-ready systems that solve real-world problems through intelligent design and robust engineering.

Full-stack FastAPI + Next.js starter, extended by a module registry

JWT auth, role-based access control, and a dual-database backend, wired up from the start. When the starter doesn't have a feature you need, install one from the registry - a page, wired to a backend route and database model, in one command.

Features:

• JWT authentication - NextAuth v5, AES-256-CBC encrypted payloads
• Role-based access control - Admin, user, and guest roles enforced on both frontend and backend
• Dual database support - SQLite by default, MongoDB with one environment variable
• Module registry - Install full-stack features (frontend routes, backend logic, database models) from the CLI, no account required
• AI Builder - Generate a module from a prompt, or publish your own for others to install
• Web and mobile, one codebase - Same frontend bundles into native Android/iOS via Capacitor
• Docker Compose stack - Frontend, backend, and an optional nginx/SSL reverse proxy, ready to run

Tech stack: Python (FastAPI, SQLAlchemy, Motor) + Next.js/TypeScript + Tailwind CSS + Docker

Repo >> | Registry >>

Self-evolving autonomous learning loop for TensorFlow RL agents

Replace your training loop with an autonomous optimization system that continuously evolves policies, tunes hyperparameters, and detects plateaus automatically. Perfect for RL problems in trading, robotics, game AI, and adaptive control.

Core capabilities:

• Autonomous training loop - Continuous stepping with no fixed episode count; system detects convergence
• Intelligent backoff scheduling - ACTIVE → COOLING → DORMANT state machine with exponential backoff
• Hyperparameter tuning - Two-phase finite difference estimation adapts parameters in real-time
• Checkpoint management - Automatic rollback to best-known policies on degradation
• Ensemble support - Run multiple agents simultaneously as weighted ensemble for non-stationary environments
• Custom agents & evaluators - Subclass for PPO, SAC, DQN, or any RL algorithm
• Live pipelines - Stream real-time data (financial markets, sensor feeds) as training data

Key algorithms:

• BackoffOptimizer — Two-phase finite difference with adaptive perturbation
• PBTOptimizer — Pseudo population-based training with history-based selection
• PolicyManager — Evolution tracking, rollback, and ensemble weighting

Perfect for: Trading bots, game AI, robot control, portfolio optimization, multi-agent systems

View >> | PyPI Package >>

Visual platform for building and orchestrating AI agents

Enterprise-grade agent orchestration without the boilerplate. Build multi-agent teams, define complex workflows, and manage evaluations entirely through a visual interface—no SDKs, no glue code.

Key capabilities:

• Visual orchestration - Drag-and-drop agent teams, sequential pipelines, parallel DAG workflows
• No vendor lock-in - Swap providers (OpenAI → Anthropic → Ollama) instantly without reconfiguring agents
• Production security - JWT auth, TOTP 2FA, AES end-to-end encryption, secrets vault, RBAC, rate limiting
• MCP-native - First-class Model Context Protocol support for external tool integration
• Self-hosted & open-source - Complete data ownership; runs entirely on your infrastructure
• Sandbox integration - Isolated containers for safe code execution within agent workflows
• Evaluation pipelines - Built-in testing and metrics for agent performance
• Human-in-the-loop - Review, refine, and approve agent decisions at every stage

Tech stack: Python (FastAPI) + React/TypeScript + PostgreSQL + Docker

View >>

Problem-first: Every project starts with a real-world constraint that existing solutions don't address.

User empathy: Tools succeed when they reduce friction. I obsess over UX for both end-users and developers.

Production focus: Systems must be deployable, maintainable, and performant at scale. Security and observability are non-negotiable.

Open by default: Code and architecture benefit from transparency. Most of my work is open-source.

Frontend: React, Next.js, TypeScript, Tailwind CSS, HTML/CSS
Backend: Python (FastAPI), Node.js, PostgreSQL, MongoDB
AI/ML: TensorFlow, Keras, Gymnasium, LangChain, LLM integrations (OpenAI, Anthropic, Ollama)
Infrastructure: Docker, Docker Compose, Celery, Redis, Server-Sent Events
Advanced: RL optimization, hyperparameter tuning, policy evolution, ensemble methods

I'm always interested in collaborating on AI systems, ML infrastructure, or interesting technical challenges.

Building systems at the intersection of AI, machine learning, and software engineering