Applied AI Engineer · Machine Learning · PhD in Deep Reinforcement Learning · Melbourne, Australia
I build AI systems that hold up in production: LLM pipelines where every model output is treated as untrusted until it passes automated evaluation gates and human review, reinforcement learning agents trained on GPU clusters, and ML on messy real-world data.
🌐 Portfolio: rasheibani.github.io · 💼 LinkedIn · ✉️ rasheibani@gmail.com
- Evaluation-gated LLM code migration (Parks Victoria): Microsoft Foundry models and GitHub Copilot CLI translating legacy Oracle PL/SQL to PostgreSQL. Retrieval over a curated knowledge base, strict JSON contracts, syntax/dependency/behavioural/security gates and human review; 775 validated objects shipped.
- Production data and MLOps on Azure: restartable migration runbooks, source-vs-target validation, guarded rollback and Azure DevOps release pipelines.
| Repo | What it is |
|---|---|
| Wayfinder-RL | Deep RL agents (PPO, DQN; PyTorch, Stable-Baselines3) that test whether natural-language route instructions can be followed, across 742 synthesised indoor environments. Published in IJGIS and JOSIS. |
| Landmark_Placement | Multi-objective integer programming (Gurobi, DEAP) to place landmarks that reduce navigation ambiguity. Published in Computers, Environment and Urban Systems (2024). |
Python PyTorch Stable-Baselines3 LLMs & agents RAG LangChain Azure AI Foundry Evals & guardrails PySpark SQL PostgreSQL/PostGIS Azure DevOps Docker Gurobi
- Deep reinforcement learning for assessing route instruction usability in complex indoor spaces, IJGIS (2025)
- Automated validation of route instructions in indoor environments, JOSIS (2025)
- Strategic allocation of landmarks to reduce uncertainty in indoor navigation, CEUS (2024)
- A new approach to flood susceptibility assessment in data-scarce and ungauged regions, Journal of Hydrology (2019, 180+ citations)
Full list on my portfolio.