LLM & Generative AI systems
RAG pipelines with hybrid retrieval and vector stores, agentic workflows with tool use, evaluation harnesses and guardrails — designed around the metrics that decide whether a system ships: latency, cost and throughput.
Open to remote AI/LLM Engineer roles worldwide
AI Engineer with a PhD in deep learning and computer vision, and five years shipping AI-enabled software in industry.
Three things I am hired for, and the evidence behind each.
RAG pipelines with hybrid retrieval and vector stores, agentic workflows with tool use, evaluation harnesses and guardrails — designed around the metrics that decide whether a system ships: latency, cost and throughput.
Object detection and image classification at research depth — a novel YOLOv5-based handwriting recognition model and hybrid transfer-learning models for medical imaging, both peer-reviewed and published.
The unglamorous half that makes AI usable: ETL and orchestration, warehouse modelling, API design, and full SDLC ownership of ERP-integrated systems running in a live manufacturing business.
A few projects that show how the research and the engineering meet.
Novel YOLOv5-based detection and recognition model for handwritten documents.
Hybrid deep transfer-learning models for medical image classification.
Retrieval-augmented, tool-using assistant over technical documents.
Peer-reviewed work in indexed journals and Springer proceedings.
Frontiers in Artificial Intelligence ESCI, Scopus
International Journal on Artificial Intelligence Tools SCI-Expanded, Scopus
ICDEC 2024 - Springer Lecture Notes in Networks and Systems Scopus
Notes on building AI systems, computer vision and data engineering.
Accuracy is not what gets a retrieval-augmented system into production. Here is the short list of numbers I…
My PhD work on handwriting detection was really a document-understanding problem. Here is how that reframes…
Research rigour and production delivery, in one person. I am open to remote ai/llm engineer roles worldwide and available for consulting through Binotix.