ML Engineer based in London with an MSc in Artificial Intelligence and Adaptive Systems from the University of Sussex. I build applied AI and machine-learning systemsβfrom data and model experiments to usable developer and product experiences. My research focused on reinforcement learning, specifically asymmetric forgetting in dual-value RL architectures for non-stationary environments.
Previously the sole Frontend Developer at MedAll, where I owned the frontend and most product-interface design for a medical learning platform. I bring both research depth and production engineering discipline to everything I build.
Currently exploring: MCP (Model Context Protocol) for ML operations Β· Multi-agent systems with LangGraph
| Evidence | What it means for a hiring team |
|---|---|
| MSc AI, University of Sussex (2025) | Research-led ML foundation, including a PyTorch reinforcement-learning dissertation. |
| Production product delivery | Sole Frontend Developer at MedAll (2020β2022): owned frontend delivery and most UI/UX design for a medical-learning product. |
| Hackathon result | 3rd place, 39 teams / 150 participants at Hack Night London (Tessl, May 2026) for CareerMentorGraph, built solo in one evening. |
| Problems I have solved | Trustworthy PDF question answering with citations; explainable graph-based career planning; no-code ML workflows; local-first AI tools and health-product prototypes. |
| Best technical evidence | LocalDocRAG Β· CareerMentorGraph Β· Crop Yield Forecasting Β· Asymmetric Forgetting RL |
Recruiter quick links: LinkedIn Β· Email Β· Featured repositories
Machine Learning & AI
Python PyTorch TensorFlow scikit-learn HuggingFace LangChain LangGraph
MLOps & Infrastructure
Docker FastAPI PostgreSQL GitHub Actions AWS GCP
Also Proficient In
TypeScript JavaScript C++ SQL Linux Git
Original research from my MSc dissertation
A dual-value reinforcement learning architecture that separates reward and punishment learning signals with asymmetric forgetting rates. Improves adaptation in non-stationary environments where optimal strategies shift over time.
PyTorch OpenAI Gym Reinforcement Learning
AI career guidance as a knowledge graph
Maps skills, gaps, and learning paths using graph reasoning and LLM-powered analysis. Surfaces personalised upskilling routes based on role targets and current skill state.
Python Knowledge Graphs LLMs
PDF Q&A with source citations, runs locally
Upload PDFs, ask questions, get page-cited answers. Runs fully offline on a Raspberry Pi. LangChain retrieval pipeline with pgvector, FastAPI backend, React frontend, Docker Compose deployment.
RAG LangChain pgvector Docker FastAPI
PyTorch MLP across 165 countries and 102 crops
Multi-source climate, soil, and land-cover data fused into a single MLP trained on 52K+ samples. R squared of 0.9452, Pearson r of 0.9681. One-year-ahead forecasts with per-country breakdown.
PyTorch MLP Climate Data scikit-learn
Asymmetric Forgetting in Dual-Q Reinforcement Learning β MSc Dissertation, University of Sussex (2025)
Developed a dual-value RL model separating reward and punishment learning signals with asymmetric forgetting to improve adaptation in non-stationary environments. Evaluated in a custom 2D grid-world with shifting reward locations. Analysed results in the context of computational forgetting, stability-plasticity trade-offs, and implications for adaptive AI systems.
π MSc Artificial Intelligence and Adaptive Systems β University of Sussex, 2025
π BSc Computer Science β Amirkabir University of Technology (Tehran Polytechnic), 2020
π Machine Learning Specialization β Stanford University & DeepLearning.AI
π CS50: Introduction to Computer Science β Harvard University
Open to ML Engineer, AI Engineer, and LLM/Agentic AI roles in London.
If you're looking for someone who builds production AI systems, not just notebooks, let's talk.