JAAI
CreAtive World
Yasek · AI Researcher & Director
PhD Candidate, Artificial Intelligence
MSc Data Science & AI (Distinction)
Director, JAAI CreAtive World Ltd
I design and build the systems behind frontier AI: reinforcement-learning environments, evaluation frameworks, and the agents and tools that put them to work. Research-grade thinking, shipped as working software.
Scroll

I work at the point where doctoral AI research meets production engineering. My research applies long-document Transformer architectures (Longformer, LED) to the systematic evaluation of AI governance policy: models that read hundred-page policy documents whole, instead of losing the thread across chunked fragments.

On the engineering side, my day-to-day toolkit runs deep: reinforcement learning end to end (custom Gymnasium environments, PPO and SAC training with Stable-Baselines3, reward shaping and difficulty calibration against expert baselines), the Hugging Face ecosystem for transformer fine-tuning and inference, retrieval pipelines with sentence embeddings and extractive QA, and the full LLM training loop: RLHF and SFT workflows, preference ranking, rubric and scaffold design, adversarial testing. Everything I build ships as working, containerised software with measured results, not notebooks that die after the demo.

Through JAAI CreAtive World Ltd, I turn that stack into applied systems for businesses: agents, evaluation tools, and data products. Several are shown below as working examples.

What I Build
Practical AI systems for businesses that need more than a chatbot wrapper — built, tested, and shipped end to end.
AI Agents & Workflow Automation
Task-planning agents that call real tools, reason step by step, and plug into your existing workflows.
Internal Knowledge Assistants
Retrieval-augmented Q&A over your own documents, with real citations and confidence scoring, kept fully private.
Brand & Market Intelligence
Real-time sentiment and monitoring dashboards that turn public data into decisions.
Custom Simulation & Reinforcement Learning
Bespoke environments and trained agents for control, optimisation, or research problems off the shelf tools can't solve.
AI Governance & Policy Analysis
NLP-driven analysis of policy and compliance documents against recognised ethical frameworks.
Model Evaluation & Alignment
Rubrics, scoring frameworks, and adversarial test design for teams that need to know their model actually works.
I currently collaborate with some of the leading AI labs and platforms on the market, contributing to how frontier AI systems are trained and evaluated. I take on a limited number of engagements at any time.
Selected Work
Working systems, built end to end — not slide-deck concepts.
PolicyCraft
NLP platform analysing university AI policy documents against UNESCO, JISC and BERA frameworks, generating evidence-based recommendations.
NLPFlaskPolicy Analysis
policycraft.jaai.co.uk →
AgentDesk
A task planner that calls its own tools step by step, with a transparent reasoning trace.
AI AgentsPythonTool Use
View project →
AutoLander
A custom 2D physics environment trained with reinforcement learning, with a live mission-replay dashboard.
Reinforcement LearningPPO
View project →
AskDocs
Local document Q&A with page-level citations. Nothing leaves the machine.
RAGTransformers
View project →
BrandPulse
Real-time brand sentiment dashboard: live news ingestion, transformer scoring, shareable reports.
NLPStreamlit
View project →
Stroke Risk Platform
Healthcare ML in two parts: a full modelling study (imbalance handling, model comparison, fairness discussion) and a patient-management web app built around it.
Healthcare MLXGBoostFlask
View project →
Currently accepting a limited number of engagements.
If your business needs AI built properly - measured, tested, and shipped - send a short brief of the problem. Serious enquiries receive a personal reply. Case studies and references available to prospective clients.
LinkedIn GitHub ORCID