Jonathan Atiene
Senior AI engineer specialising in production LLM and agentic systems.
United Kingdom 🇬🇧
I architect and ship agent harnesses, RAG and reranking pipelines, context and memory systems, evaluation frameworks, structured outputs, tool orchestration and human-in-the-loop workflows. My work spans Llama fine-tuning and deep-learning research through to Python and TypeScript platforms on AWS.
Deep technical skills
LLM and agent architecture
Production agent harnesses, multi-agent workflows, LangGraph state graphs, tool orchestration, Pydantic-validated structured outputs, context windows, long-term memory, human approval gates and failure recovery.
Retrieval and knowledge systems
RAG pipelines, vector retrieval with Qdrant, query and document representation, reranking, citation grounding and knowledge graphs—including retrieval across more than 70,000 clinical records.
Language models and fine-tuning
Llama 3B fine-tuning with PyTorch, Hugging Face and Vertex AI; dataset preparation, synthetic-data pipelines, prompt and context optimisation, and open-model inference workflows.
Deep learning and model efficiency
RAAGR²-Net brain-tumour segmentation, model pruning, weight sharing and parameter-efficiency research focused on reducing model size and deployment cost without losing useful performance.
Evaluation and reliability
Agent evaluation, red teaming, Langfuse tracing, schema validation, evidence checks, observability and recovery paths for long-running model and tool workflows.
AI platform engineering
Python, TypeScript, Node.js, PostgreSQL, AWS Glue, serverless AWS services, GCP, Docker, Terraform, CI/CD and scalable data and ML pipelines.
Featured writing
Introducing Docket AI: Invoices, Quotes and Receipts from WhatsApp
I built Docket AI so small businesses can create branded invoices, quotes, receipts, agreements and delivery notes by sending a WhatsApp message.
AI Engineering · 3 min
- The End of AI Slop: Engineering a Voice for AI-Assisted WritingAI Engineering
- Playing on "peak" as mountain and "to see" as beholdingWriting
- Ship Your Own AI: I Built a Desktop App That Turns Any Dataset into a Fine-Tuned ModelAI Engineering
Selected systems
My strongest work sits in the engineering around the model: context, memory, retrieval, tools, evaluation, infrastructure and human review.
Docket AI
Business documents from a WhatsApp message
A WhatsApp-native agent that turns plain messages into branded invoices, quotes, receipts, agreements and delivery notes, asking for missing details before it generates anything.
HiveScience AI
Audience intelligence at enterprise scale
A patent-pending ML environment for Fortune 500 teams, with a production agent harness spanning context, memory, tool orchestration, structured outputs, validation and failure recovery.
Axon · Neurochecklists
Evidence-grounded answers for neurology
A privacy-conscious RAG agent that searches and reranks more than 70,000 clinical checklist records to return structured, cited answers in under 2.5 seconds.
AskHook
Social intelligence from short-form video
An agentic system that distinguishes organic virality from paid promotion and turns patterns across more than 500 content pieces into actionable ideas in minutes.
Currently exploring
- Agentic memory
- Open-model fine-tuning
- AI system reliability
- Agent evaluation
- Context engineering
Essays in your inbox
Occasional writing about intelligent systems, agents and the engineering behind them.