AI Engineer · Backend & Quantitative Systems
I design and ship production AI systems: the models, the services around them, and the infrastructure that keeps both standing under load. My work sits where model behaviour meets hard system constraints such as latency budgets, cost ceilings and failure modes that have to be designed for rather than discovered.
I build AI powered products and client systems at Hashed System FZCO, and work independently with teams that need AI capability architected properly the first time rather than retrofitted later.
| Domain | What I do |
|---|---|
| Applied AI | LLM orchestration, agentic workflows, RAG architectures, tool and MCP integrations, evaluation harnesses |
| Machine Learning | Supervised and unsupervised modelling, feature engineering, validation design, full lifecycle from prototype to serving |
| Deep Learning | Neural architecture design, training and fine tuning, sequence and transformer models, inference optimisation |
| Explainable AI | SHAP attribution, sparse autoencoders, residual stream analysis, interpretability tooling for opaque models |
| Time Series Forecasting | Forecasting model design, regime detection, temporal feature engineering, walk forward validation |
| Quantitative Systems | High frequency and low latency execution paths, backtesting infrastructure, market data pipelines, deterministic risk gating |
| Backend Engineering | API and service architecture, async and background workloads, caching and queueing, data modelling |
| System Design | Distributed architecture, scalability and throughput planning, fault tolerance, observability and monitoring |
| Layer | Stack |
|---|---|
| Core Language | Python (primary), with JavaScript, Java, C++ and SQL |
| AI / ML | PyTorch, TensorFlow, Keras, scikit-learn, Transformers, SHAP |
| LLM Stack | RAG pipelines, vector databases, agent frameworks, MCP, evaluation tooling, model serving |
| Backend | FastAPI, Django, Django REST Framework, REST and WebSocket APIs |
| Data | PostgreSQL, MongoDB, MySQL |
| Quant Tooling | NumPy, pandas, backtesting engines, market data ingestion, statistical validation |
| Infrastructure | Docker, AWS, Linux, CI/CD, monitoring and logging |
| Principle | In practice |
|---|---|
| Constraints before code | Latency, cost and failure behaviour are decided at design time, not discovered in production |
| Boring where it counts | Novel architecture only where the problem demands it, proven patterns everywhere else |
| Ownership end to end | Architecture, implementation, deployment and the monitoring that proves it works |
| Systems outlive engineers | Clear boundaries, readable data flow and documentation that survives handover |
| linkedin.com/in/mehar-umar | |
| me.umar0027@gmail.com |
Open to conversations about AI engineering, system design and quantitative infrastructure.