AI/ML & LLM Engineer: building ML and LLM systems that are measured, not guessed.
I'm an AI/ML & LLM engineer building classical ML models, RAG pipelines, multi-agent systems and the eval harnesses that keep them honest. My approach was shaped by work in healthcare AI, where a confident wrong answer is a real risk and evaluation isn't optional.
- ML: supervised modelling, gradient-boosted ensembles, feature engineering, model selection
- LLMs: retrieval-augmented generation, tool-using agents, LLM-backed APIs
- Measure: cross-validation, golden datasets, RAG metrics, regression tests, tracing
- Currently: multi-agent systems, grounded RAG, and LLM fine-tuning & evaluation
- Open to: remote AI/ML & LLM Engineering roles and open-source collaborations
Languages & Tools
Machine Learning
LLM & RAG
Evaluation & Observability
| Project | What it does | Stack |
|---|---|---|
| PostCareAI | Multimodal multi-agent system for clinician-supervised post-operative patient follow-up | Agents · Multimodal LLMs |
| ActLens | Citation-backed EU AI Act compliance assistant | LangChain · Hybrid RAG |
| clinicdoc | Audits clinical NLP datasets before training: offline, deterministic, no cloud model | Python · Data quality |
| Surgical Tool Navigation | Real-time ArUco marker tracking with AR guidance (direction + distance) | OpenCV · Computer vision |
- Eval first. No prompt, retriever or model change ships without a metric that moved.
- Ground every answer. If it can't cite its source, it doesn't get said.
- Reproducible by default. Pinned data, versioned prompts, deterministic test sets.
- Humans stay in the loop. Where mistakes are costly, the system assists; experts decide.