Basel M.

👋 Hello, I'm Basel (yes, the Swiss city was named after me). I'm a PhD researcher in machine learning and operations research (doctoral work conducted at TU Delft in residence at the Austrian Institute of Technology, Vienna, thesis submitted, defense scheduled for January 2027), working in this field since my Master's in 2019. Based in Vienna. Outside of research, I compose and orchestrate music. I build systems that have to survive contact with reality, not just demos: agentic AI pipelines that ground every claim in real evidence, full-stack web and mobile applications, and quantitative trading systems that trade real capital. My doctoral research applies the same rigor, mathematical optimization and decision-making under uncertainty, to power-systems operation, but the freelance work below is where I build for clients directly. 🚀 𝐒𝐞𝐥𝐞𝐜𝐭𝐞𝐝 𝐏𝐫𝐨𝐣𝐞𝐜𝐭𝐬 🛠️ 𝐂𝐥𝐢𝐧𝐭𝐞𝐥 — built solo for one of Austria's largest banks (name withheld, NDA): an agentic pipeline that turns hundreds of pages of prospect filings into a fully cited briefing, every figure traceable to a source page. Measured, not just demoed: precision 1.000, coverage 0.879 over 51 hand-labeled rows. FastAPI, PostgreSQL, Redis, React/TypeScript, single-command Docker deploy. 𝐒𝐜𝐡𝐨𝐥𝐚𝐫𝐃𝐫𝐚𝐟𝐭 — a hybrid-RAG research workbench: dense and sparse retrieval, a LangGraph agent for open-ended literature exploration, and a citation-grounded generation engine that verifies every AI-drafted citation against real evidence before it's shown. ~34,000 lines across a FastAPI/LangGraph backend and desktop (Tauri), web, and terminal frontends. 𝐂𝐚𝐫𝐞𝐞𝐫𝐢𝐟𝐲 — a self-hosted CV manager with a two-step agentic editing flow: one step analyzes a job description against your CV and proposes edits, a second applies only the ones you accept, both constrained by explicit anti-fabrication rules so it can reframe existing experience but never invent it. 𝐐𝐓𝐒 — a from-scratch reinforcement-learning trading framework delivered over about a year to a US-based hedge fund client (name withheld, NDA): four RL algorithms (SAC, PPO, QR-DQN, DDPG) implemented directly in raw PyTorch, a two-tier labeling and pretraining system, live broker integration, and a real-time monitoring dashboard, improving the client's strategy performance 20 to 40% versus baseline. 𝐁𝐚𝐜𝐤𝐒𝐭𝐮𝐝𝐢𝐨 — open-source (MIT) on GitHub: one YAML file becomes a full production-shaped FastAPI backend, auth, RBAC, row-level security, async support included. Its own test suite proves the generated code actually runs, not just that templates render. CI green on every push. 𝐓𝐫𝐚𝐜𝐤 𝐑𝐞𝐜𝐨𝐫𝐝 📈 Top Rated Plus on Upwork with a 100% Job Success Score across roughly 7 client engagements in agentic AI, trading, analytics, and automation. 𝐓𝐞𝐜𝐡𝐧𝐢𝐜𝐚𝐥 𝐓𝐨𝐨𝐥𝐤𝐢𝐭 🧰 𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐀𝐈 & 𝐃𝐚𝐭𝐚 𝐒𝐜𝐢𝐞𝐧𝐜𝐞: LangChain, LangGraph, LangSmith, MCP, retrieval-augmented generation (hybrid dense/sparse, reranking), Qdrant, Neo4j, PyTorch, TensorFlow, Optuna, pandas, statistical hypothesis testing and time-series forecasting 𝐖𝐞𝐛 & 𝐌𝐨𝐛𝐢𝐥𝐞 𝐃𝐞𝐯𝐞𝐥𝐨𝐩𝐦𝐞𝐧𝐭: FastAPI, SQLAlchemy (sync/async), PostgreSQL, Redis, React, TypeScript, React Native, Tauri, Docker, JWT/RBAC/row-level security design 𝐐𝐮𝐚𝐧𝐭𝐢𝐭𝐚𝐭𝐢𝐯𝐞 𝐓𝐫𝐚𝐝𝐢𝐧𝐠 & 𝐎𝐩𝐞𝐫𝐚𝐭𝐢𝐨𝐧𝐬 𝐑𝐞𝐬𝐞𝐚𝐫𝐜𝐡: reinforcement learning built from scratch (SAC, PPO, DDPG, QR-DQN), MetaTrader 5 live execution, mixed-integer and nonlinear programming, SOCP/SDP relaxations, Benders decomposition, column-and-constraint generation, scheduling problems (vehicle routing, machine scheduling, unit commitment) 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞𝐬 & 𝐓𝐨𝐨𝐥𝐬: Python, Julia (JuMP.jl), TypeScript, C/C++, MATLAB 𝐖𝐡𝐲 𝐖𝐨𝐫𝐤 𝐖𝐢𝐭𝐡 𝐌𝐞 ✅ 1️⃣ 𝐁𝐮𝐢𝐥𝐭, 𝐧𝐨𝐭 𝐚𝐬𝐬𝐞𝐦𝐛𝐥𝐞𝐝: my RL algorithms and agentic pipelines are implemented from scratch, not stitched together from off-the-shelf libraries, so I can actually explain and adapt every part of the system, not just configure it. 2️⃣ 𝐑𝐢𝐠𝐨𝐫 𝐛𝐲 𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠: a PhD spent on methods that have to hold up under review means I default to real baselines, honest evaluation, and telling you when a result is preliminary rather than overselling it. 3️⃣ 𝐀 𝐭𝐫𝐚𝐜𝐤 𝐫𝐞𝐜𝐨𝐫𝐝 𝐲𝐨𝐮 𝐜𝐚𝐧 𝐜𝐡𝐞𝐜𝐤: Top Rated Plus, 100% Job Success Score, and engagements that shipped and stayed shipped, not just one-off proofs of concept. I'm open to new engagements in agentic AI, data science, full-stack web/mobile development, and quantitative trading, whether that's a well-defined build or an open problem you're still framing. Feel free to reach out to discuss your project. 💬