Senior AI Engineer working on production agent systems, evaluation, LLMs and applied ML.

I like building things, taking them apart, and understanding what is happening underneath the abstraction — especially when that means turning research ideas into systems that actually run.

• 🤖 Agentic AI — production agents, tool use, context engineering, orchestration, evaluation and reliability
• 🧪 AI Evaluation — behavioural testing, LLM-as-a-judge reliability, regression detection and evaluation infrastructure
• 🧠 LLMs & Transformers — fine-tuning, embeddings, NLP, model architecture and understanding models from the inside out
• 👁️ Computer Vision — image understanding, detection, OCR and vision-language models
• ⚙️ AI Engineering — APIs, data pipelines, retrieval systems and production infrastructure around models
• 🔬 ML research & experiments — implementing papers, testing unusual ideas and exploring new model architectures

A few public projects that represent what I'm interested in right now:

🤖 Meta-Eval — An evaluation harness exploring behavioural invariance and the reliability of LLM-as-a-judge systems. It supports running open models locally, comparing multiple judges and separating model generation from re-judging so evaluations can be repeated without regenerating outputs.

🧪 LLM Fine-tune & Eval — A portfolio-scale fine-tuning and evaluation harness built around Qwen2.5-3B-Instruct, LoRA/QLoRA and repeatable evaluation. The focus is deliberately on the system around the model: experiment tracking, metrics, regression detection and serving.

🛡️ DiffVax — Reimplementation and experimentation around optimization-free image immunization against diffusion-based editing. Uses a trained Nested U-Net to generate perturbations in a single forward pass rather than optimizing every image individually.

🧠 ELECTRA PyTorch — A PyTorch implementation of ELECTRA with configurable generator/discriminator architectures, mixed-precision training, gradient accumulation, checkpointing and hyperparameter optimisation.

📊 AI Usage Dashboard — A self-hosted dashboard for monitoring AI API spend and token usage across OpenAI and Anthropic, with SQLite persistence, project filtering and Docker deployment.

You'll also find smaller experiments around Transformers, graph ML, translation, model understanding and ML tooling.

Python · PyTorch · Hugging Face Transformers · LangChain · LangGraph · LangSmith · FastAPI · MLflow · Elasticsearch · Docker · Git

Increasingly:

Local LLMs · VLMs · MoE · RAG · Agentic AI · Model evaluation

I'm a BSc Software Engineering graduate from the University of Portsmouth. My career has moved from software engineering and industrial automation into computer vision, ML engineering and production AI systems.

I like hard technical problems, weird side projects, learning how things work under the hood, and turning half-formed ideas into something that actually runs.

I'm also, unfortunately, a chronic ideas person.

📐 BSc Software Engineering · University of Portsmouth
🧠 AI / ML · Agents · Evaluation · Transformers · Computer Vision
📬 sam.bartlett858@gmail.com