Writing
Writing and research
Notes from building and evaluating systems with knowledge graphs and LLMs. Most of my articles are published on Medium.
Articles
-
Evaluating a Fine-Tuned LLM Using a Knowledge Graph and Ground Truth Data
Using a travel knowledge graph built from Geonames, Wikidata, TripAdvisor and Yelp as ground truth to evaluate how well a fine-tuned model understands real places.
LLM evaluation
-
Why I Built Tripoh — A Real-Time Travel Assistant That Understands You
The thinking behind Tripoh, from a graph-based travel assistant to one built on large language models.
Product
-
Revolutionizing Social Media Data Analysis: How Gemini 2.5 Makes Relation Extraction and NER Effortless
Using Gemini 2.5 for named entity recognition and relation extraction on informal, multilingual social media text.
Information extraction
-
Building a Graph-Transformer Hybrid: A New Approach to Language Models
An experimental architecture that adds entity-aware graph attention to a transformer.
Graph machine learning
-
Beyond Initial Representations: Why Reasoning Layers Are Essential in Graph Embedding
Why raw node embeddings miss indirect relationships and structure, and how reasoning layers refine them.
Graph machine learning
-
From 10,000 Nodes to 10 Million: Our Journey Scaling a Tourism Knowledge Graph
Lessons from growing a Neo4j tourism knowledge graph by three orders of magnitude while keeping queries fast and memory under control.
Knowledge graphs
-
Graph Embedding Pitfalls: Overcoming Performance Bottlenecks in Graph Transformation
What slowed down a pipeline that read a graph, computed TransB embeddings and wrote them back, and how to work around it.
Knowledge graphs
Research
-
Reduce False Negative in Distant Supervised Learning Using Dependency Tree-LSTM to Construct a Knowledge Graph
Women in Machine Learning (WiML) Workshop at NeurIPS 2022, 2022. WiML 2022 programme
Research interests
- Distant supervision for relation extraction
- Neural relation extraction
- Prompt learning
- Distributed knowledge graphs