Patteera Triamamornwooth

I build and evaluate RAG systems, with a focus on the part most teams skip: measuring whether retrieval actually works. At Fufong Labs I built our evaluation pipeline on RAGAs and created the custom QA benchmark dataset behind it, which became the reference we validate every chatbot iteration against. I lifted retrieval performance 10–20% with a reranking layer and 50% by switching to multilingual embeddings for cross-lingual matching, and I'm researching deterministic IR-based metrics as a cheaper alternative to LLM-as-judge scoring.

Before that I evaluated LLMs across 10 industry-standard benchmarks with OpenThaiGPT and curated Thai benchmark datasets including HellaSwag-TH, cited in the FineWeb2 dataset survey. Earlier, at iBOTNOI, I fine-tuned an Arabic–English translation model above baseline BLEU and shipped a text classifier at 0.80 accuracy.

MSc in Data Science & AI (University of Liverpool), with working experience across classical ML, NLP and computer vision alongside LLM work.