I am Melek Moalla, an AI engineering student at ESPRIT in Tunisia, building projects in machine learning, deep learning, computer vision, reinforcement learning, and software engineering.
I build intelligent software systems that connect machine learning, backend engineering, and product-facing interfaces into one coherent experience.
- 🧠 Applied ML & deep learning — plant genomics, disease detection, explainable surrogate models
- 👁️ Computer vision — real-time hand tracking, gesture recognition and AR interaction
- ⚡ Real-time systems — WebSocket streaming, Redis, observability dashboards
- 🧩 Full-stack AI products — FastAPI / Django backends with React and Flutter frontends
- 📚 RAG & NLP — knowledge-grounded assistants and workflow automation
- 🎮 Reinforcement learning — PPO agents, reward design and honest evaluation of what agents actually learn
- 🤖 Agent tooling — orchestrating Claude Code and Codex CLI agents side by side
I care about systems that are technically credible and actually usable: clean APIs, interpretable ML, and interfaces that make complex outputs feel actionable.
| Project | What it does | Stack |
|---|---|---|
| Reward-Goblin | Reward hacking made visible: real PPO agents exploit misspecified reward functions, measured against a separate true objective across seeds and unseen layouts | Python, PyTorch, Stable-Baselines3, Gymnasium, Pymunk, FastAPI |
| Hydra-Agent-Orchestrator | Cross-platform desktop app that runs, monitors and coordinates Claude Code and OpenAI Codex CLI agents in parallel | TypeScript, Node.js, MCP |
| Reinforcement-Learning-Memory-Dungeon | PPO lab comparing feed-forward, short-history and recurrent agents in partially observable dungeons, with memory-deletion experiments | Python, PyTorch, Streamlit |
| Vector-CS | Controls the real Windows desktop with webcam hand gestures: grab, throw, snap and resize windows | Python, Computer Vision, Win32 API |
| Deep-Learning-Plant-DNA-Optimization | Plant disease detection, crop DNA optimization, IoT agronomy analytics and genomic decision support | Python, Deep Learning, FastAPI, Flutter |
| AI-Driven-Therapeutic-Protein-Plant-Synthesis | Plant-host-aware protein DNA optimization with grouped surrogate modeling, explainability and decision dashboards | Python, Streamlit, scikit-learn, XGBoost, SHAP |
| Real-Time-Payment-Observability-Dashboard | Privacy-safe live payment monitoring with synthetic replay and guarded analytics | FastAPI, React, TypeScript, WebSockets, Redis |
| Telekinesis-CV | Hand-controlled AR playground with gesture recognition and custom physics | Python, OpenCV, MediaPipe |
| Festy-Event | Event booking platform with QR tickets, venue maps, availability calendar, PDF invoices and admin analytics | Django, Leaflet, Chart.js, WeasyPrint |
| LogiXpress-WebSite | Logistics platform with delivery tracking, packaging workflows, notifications and AI-assisted modules | PHP, MySQL, Firebase, Python |
| Area | Tools |
|---|---|
| AI / ML | PyTorch, TensorFlow, scikit-learn, XGBoost, SHAP, pandas, NumPy |
| Reinforcement Learning | Stable-Baselines3 (PPO, A2C), Gymnasium, Pymunk |
| Computer Vision | OpenCV, MediaPipe |
| AI Product Patterns | RAG, NLP workflows, explainable ML, analytics dashboards |
| Backend | FastAPI, Django, REST APIs, WebSockets, Redis |
| Frontend & Mobile | React, TypeScript, JavaScript, Flutter |
| Data & Storage | PostgreSQL, MySQL, Firebase |
| Delivery | Git, GitHub Actions, Docker |
Stats, language and project cards, and the profile views badge are refreshed daily by a GitHub Action. The views badge keeps its last good value if the counter is unavailable.
- Currently improving Reward Goblin, my reinforcement-learning playground for exploring reward hacking with real PPO agents and separate task-success evaluation
- Smarter RAG-driven experiences with strong retrieval quality and practical UX
- AI-first full-stack applications where backend intelligence directly shapes the interface
- Real-time computer vision and streaming systems that run reliably outside the notebook
- NLP and HR Tech workflows that automate real business decisions
- Evaluation tooling that checks what AI agents actually do, not just the score they report
