Thierry Damiba

Developer relations engineer at Qdrant. Author of the five-part Fine-Tuning Sparse Embeddings for E-Commerce Search series (March 2026), which fine-tunes SPLADE on the Amazon ESCI Dataset, benchmarks it against BM25, tests cross-catalog transfer, and packages the pipeline as qdrant-sparse-finetune.

The series is notable for reporting its negative results as prominently as its positive ones — the ESCI-tuned model loses to off-the-shelf SPLADE on Home Depot data and collapses on MS MARCO.

Articles

• Fine-Tuning Sparse Embeddings for E-Commerce Search — Parts 1–5

Topics

• Learned Sparse Retrieval · SPLADE · Hard Negative Mining · Embedding Fine-tuning

Related

• Qdrant — employer
• Evgeniya Sukhodolskaya — colleague; presented this work at MICES (Evgeniya Sukhodolskaya - Fine-Tuning Sparse Neural Retrievers for E-Commerce)
• qdrant-sparse-finetune — the framework the series produced